DISS. ETH No. 22490
Creating Markets for Energy Innovations
Case Studies on Policy Design and Impact
A thesis submitted to attain the degree of
DOCTOR OF SCIENCES of ETH ZURICH
(Dr. sc. ETH Zurich)
presented by Jörn Torsten Hünteler
Dipl.-Wirt. Ing., RWTH Aachen University M.Sc., Tsinghua University
born on 30.09.1984 citizen of Germany
accepted on the recommendation of
Examiner: Prof. Dr. Volker H. Hoffmann Co-examiner: Prof. Dr. Joanna Lewis Co-examiner: Prof. Dr. Stefano Brusoni
Zurich, 2015
Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
Table of Contents
ACKNOWLEDGEMENTS ...... X
ABSTRACT ...... XII
ZUSAMMENFASSUNG ...... XV
SYNOPSIS ...... 1
1. INTRODUCTION ...... 2 1.1. Mitigating Climate Change: A Mammoth Technological Challenge ...... 2 1.2. In Search for a Technological Revolution in the Energy Sector ...... 4
2. POLICY RELEVANCE ...... 6 2.1. A Growing Role for Government in Energy Innovation ...... 6 2.2. Shifting Priorities: From RD&D to Market Creation ...... 8
3. RESEARCH DESIGN ...... 12 3.1. Overarching Research Questions ...... 12 3.2. Theoretical Perspective ...... 15 3.3. Contributions to the Literature ...... 18 3.4. Research Framework and Contributions of Individual Papers ...... 19 3.5. Methodology ...... 22
4. SUMMARY OF THE PAPERS’ FINDINGS ...... 22 4.1. Paper 1: How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology ...... 23 4.2. Paper 2: Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation ...... 24 4.3. Paper 3: The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries ...... 26 4.4. Paper 4: Compulsive Policy-Making – The Evolution of the German Feed-in Tariff System for Solar Photovoltaic Power ...... 28 4.5. Paper 5: International Support for Feed-in Tariffs in Developing Countries – A Review and Analysis of Proposed Mechanisms ...... 30 4.6. Paper 6: Japan’s Post-Fukushima Challenge – Implications from the German Experience on Renewable Energy Policy ...... 32
5. CONCLUSIONS ...... 33 5.1. Tailoring Deployment Policies to Specific Energy Technologies ...... 34 5.2. Designing Deployment Policies that are Effective in the Long Run ...... 38
6. OVERVIEW OF THE PAPERS ...... 40
iii
REFERENCES ...... 41
PAPER 1: HOW A PRODUCT’S DESIGN HIERARCHY SHAPES THE EVOLUTION OF TECHNOLOGICAL KNOWLEDGE – EVIDENCE FROM PATENT-CITATION NETWORKS IN WIND POWER...... 51
1. INTRODUCTION ...... 54
2. THEORETICAL PERSPECTIVE ...... 56 2.1. The Sequential Pattern of Innovation in Systemic Artifacts ...... 56 2.2. The Influence of the Design Hierarchy on the Evolution of Artifacts ...... 57 2.3. The Influence of the Design Hierarchy on the Evolution of Knowledge ...... 59
3. RESEARCH CASE ...... 61 3.1. Rationale for Case Selection ...... 61 3.2. Scope of Analysis ...... 61 3.3. Complex Product Architecture ...... 62 3.4. Stable Service Characteristics ...... 64 3.5. Technological Change along one Trajectory ...... 64
4. DATA AND METHODOLOGY ...... 66 4.1. Empirical Strategy ...... 66 4.2. Design Hierarchy ...... 67 4.3. Patent and Patent Citation Data...... 68 4.4. Patent-Citation Network Analysis ...... 70 4.5. Patent-Content Analysis ...... 72
5. RESULTS ...... 74 5.1. Design Hierarchy ...... 74 5.2. Gradual Stabilization of Knowledge Trajectory ...... 75 5.3. Foundations of Today’s Trajectory of Knowledge Generation ...... 77 5.4. Influence of Network-External Knowledge along Trajectory ...... 79
6. DISCUSSION ...... 80 6.1. Creative Sequences in the Evolution of an Industry’s Knowledge Base ...... 80 6.2. Implications for Technology Strategy and Public Policy ...... 82 6.3. Interaction of Artifact and Knowledge Dimensions along Technological Trajectories ...... 84 6.4. Limitations and Future Research ...... 85
7. CONCLUSION ...... 86
ACKNOWLEDGEMENTS ...... 87
REFERENCES ...... 88
APPENDIX ...... 94
iv Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
PAPER 2: TECHNOLOGY LIFE-CYCLES IN THE ENERGY SECTOR – TECHNOLOGICAL CHARACTERISTICS AND THE ROLE OF DEPLOYMENT FOR INNOVATION ...... 103
1. INTRODUCTION ...... 106
2. THEORETICAL PERSPECTIVE AND LITERATURE REVIEW ...... 108 2.1. Two Contrasting Models of the Technology Life-Cycle ...... 108 2.2. Technological Characteristics and Life-Cycle Patterns ...... 111
3. RESEARCH CASES ...... 113 3.1. Characteristics of the Case Technologies ...... 113 3.2. Dominant Designs and Technological Trajectories in Solar PV and Wind Power ...... 115
4. DATA AND METHODOLOGY ...... 119 4.1. Empirical Strategy ...... 119 4.2. Patent Data ...... 120 4.3. Connectivity Analysis ...... 122 4.4. Patent-Content Analysis ...... 124
5. RESULTS ...... 126 5.1. Characterizing the Current Life-Cycle Stage ...... 127 5.2. Characterizing Previous Stages of the Technology Life-Cycle ...... 129
6. DISCUSSION ...... 134 6.1. Technology Life-Cycles in Energy Technologies ...... 134 6.2. Implications for Technology Policy ...... 136 6.3. Reconciling Empirical Evidence ...... 138 6.4. Limitations and Further Research ...... 140
7. CONCLUSION ...... 141
ACKNOWLEDGEMENTS ...... 143
REFERENCES ...... 144
PAPER 3: THE EFFECT OF LOCAL AND GLOBAL LEARNING ON THE COST OF RENEWABLE ENERGY IN DEVELOPING COUNTRIES ...... 159
1. INTRODUCTION ...... 162
2. LOCAL AND GLOBAL TECHNOLOGICAL LEARNING ...... 164 2.1. Technological Learning in Developing Countries ...... 164 2.2. Local and Global Learning Effects in Value Chains ...... 165
3. THE CASE OF THAILAND’S ELECTRICITY SECTOR ...... 167 3.1. Case Selection ...... 167 3.2. Trends and Challenges ...... 167 3.3. Targets and Support for Renewable Energy ...... 169
v
3.4. Local and Global Learning in Renewable Energy in Thailand ...... 171
4. MATERIALS AND METHODS ...... 172 4.1. General Model Framework ...... 172 4.2. Cost of Renewable Electricity Generation ...... 174 4.3. Model Specification and Scenarios ...... 177
5. RESULTS ...... 179 5.1. Impact on the Electricity Mix ...... 179 5.2. Incremental Policy Cost and Effects of Learning ...... 180 5.3. Differences between Technologies ...... 182
6. DISCUSSION ...... 183 6.1. How to Tap Local Learning Potentials ...... 183 6.2. Implications for Domestic Policy ...... 185 6.3. Implications for International Technology Support Mechanisms ...... 186 6.4. Limitations ...... 187
7. CONCLUSION ...... 188
REFERENCES ...... 189
APPENDIX A: COST OF AVOIDED ELECTRICITY ...... 196
APPENDIX B: MODEL INPUT ASSUMPTIONS ...... 198
PAPER 4: THE EVOLUTION OF THE GERMAN FEED-IN TARIFF SYSTEM FOR SOLAR PHOTOVOLTAIC POWER ...... 201
1. INTRODUCTION ...... 204
2. THEORETICAL PERSPECTIVE ...... 206 2.1. Innovation Systems Analysis as a Means to Inform Policy Interventions ...... 206 2.2. Potential Mechanisms Shaping the Dynamics of Policy Interventions in Innovation Systems .... 207
3. RESEARCH CASE ...... 209
4. METHOD ...... 210
5. EVOLUTION OF THE GERMAN FIT SYSTEM FOR PV ...... 215 5.1. Phase 1: Establishing a Sufficient Financial Incentive (until 2000) ...... 216 5.2. Phase 2: Removing Barriers to Market Growth (2000-2004) ...... 217 5.3. Phase 3: Limiting Rising Costs for Society (2004-2011) ...... 219 5.4. Phase 4: Ensuring Seamless Integration into the Market and the Electricity Grid (since 2011) . 224
6. DISCUSSION ...... 226 6.1. The German FIT as an Example of Policy Learning ...... 226 6.2. Policy-Induced Technological Change as a Driver of Policy Dynamics...... 227 6.1. Framework ‘Compulsive Policy-Making’ ...... 232
vi Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
6.2. Innovation Systems and Policy Learning: Towards an Integrated Framework ...... 236
7. LIMITATIONS AND FUTURE RESEARCH ...... 237
8. CONCLUSION ...... 238
REFERENCES ...... 240
APPENDIX ...... 245
PAPER 5: INTERNATIONAL SUPPORT FOR FEED-IN TARIFFS IN DEVELOPING COUNTRIES – A REVIEW AND ANALYSIS OF PROPOSED MECHANISMS ...... 251
1. INTRODUCTION ...... 253
2. INTERNATIONAL SUPPORT MECHANISMS FOR FEED-IN TARIFFS IN DEVELOPING COUNTRIES ... 255 2.1. Scope of Analysis ...... 255 2.2. Proposed Mechanisms to Cover FIT Cost in Developing Countries ...... 255 2.3. Three Ways to Balance the FIT Budget ...... 260
3. THAILAND’S ALTERNATIVE ENERGY DEVELOPMENT PLAN ...... 263 3.1. Electricity Sector Background ...... 263 3.2. Renewable Energy Targets and Current Policy Support ...... 264
4. METHODOLOGY ...... 265 4.1. Framework of Analysis ...... 265 4.2. Cost of FIT Payments ...... 266 4.3. Avoided Cost ...... 266 4.4. Avoided Cost Scenarios ...... 268
5. RESULTS ...... 269 5.1. Incremental Cost of and Cost Drivers of FIT Policy ...... 269 5.2. Uncertainty of Incremental Cost ...... 271 5.3. Impact of Uncertainty Under Different Supported-FIT Designs ...... 273
6. DISCUSSION ...... 274 6.1. Incremental Cost of Supported FITs ...... 274 6.2. Policy Implications ...... 276 6.3. Limitations ...... 279
7. CONCLUSION ...... 280
REFERENCES ...... 283
APPENDIX ...... 289
PAPER 6: JAPAN’S POST-FUKUSHIMA CHALLENGE - IMPLICATIONS FROM THE GERMAN EXPERIENCE ON RENEWABLE ENERGY POLICY ...... 293
1. THE JAPANESE ENERGY CRISIS AFTER FUKUSHIMA ...... 295
2. RENEWABLE ENERGY POLICY IN JAPAN BEFORE FUKUSHIMA ...... 296
vii
3. A NEW POLICY APPROACH EMERGING IN RESPONSE TO THE CRISIS ...... 297
4. LESSONS FROM THE PV FIT IN GERMANY APPLIED TO THE JAPANESE CONTEXT ...... 299
5. IMPLICATIONS FOR JAPANESE ENERGY POLICY AND A RESEARCH AGENDA ...... 303
REFERENCES ...... 305
CV ...... 311
viii Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
“Geez, Lisa, looks like tomorrow I’ll be shoveling 10 feet of global warming.”
Homer Simpson
ix
Acknowledgements
This dissertation is the output of a four-year learning process. There are many people without whom this learning process would not have been possible. My understanding of the issued I studied and the papers that came out of the research were shaped and polished on the way to their present form through feedback, questions and ideas of my supervisors, colleagues and students. Knowing that this list can only be incomplete, I want to extend my gratitude and thankfulness to the following individuals:
Volker Hoffmann, for incredible support over the last four years. I would not have started – let alone finished – this PhD thesis without your patience, continued believe in the relevance of my work, feedback at the right time, and the great working environment that you have created at SusTec in Zurich.
Joanna Lewis and Stefano Brusoni, for kindly agreeing to be on my examination committee. I greatly appreciate the time you take to read and comment on my thesis.
Florian von Wangenheim, for chairing my dissertation defense.
Venkatesh Narayanamurti, Laura Diaz Anadon and Henry Lee, for giving me the unique opportunity to spend the last year of my PhD in the Science, Technology and Public Policy program at the Harvard Kennedy School.
Norichika Kanie and the TiROP program, for making it possible for me to spend three great months at the Tokyo Institute of Technology.
Tobias Schmidt, for friendship and careful guidance, for patient supervision, and for being an inspiration throughout the three years. Your excellent advice gave this dissertation invaluable momentum.
Joern Hoppmann, for being the best office mate one could wish for of over the years. My thinking has benefited from our uncounted discussions more than from anything else (and my sleep from your couch when I was homeless in Zurich).
My co-authors, Bastien Girod, Volker Hoffmann, Joern Hoppmann, Norichika Kanie, Christian Niebuhr, Jan Ossenbrink and Tobias Schmidt, for many lively debates, productive problem-solving sessions, and fruitful outcomes. It was a great experience to collaborate with you.
x Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
My colleagues in Zurich, Tokyo and Cambridge, who gave me feedback and advice, and who made me enjoy going to work every morning in Zurich, Tokyo and Cambridge over the four last years: David Antons, Kathy Araujo, Benedikt Battke, Catharina Bening, Nicola Blum, Gabe Chan, Jérémie Coquoz, Xia Di, Balint Dioszegi, Claudia Dobringer, Ryan Ellis, Sabine Erlinghagen, Siegfried Flohr, Fanny Frei, Sonja Förster, Motoko Fujii, Bastien Girod, Suzanne Greene, Saya Goto, Tobias Griesshaber, David Grosspietsch, Daisuke Hayashi, Monica Heinz, Masahiko Iguchi, Erin Kennedy, Julian Koelbel, Rina Kojima, Maki Koga, Tillmann Lang, Alexander Langguth, Nils Lehmann, Alexandru Luta, Xu Lei, Tijs van Maasakkers, Jochen Markard, Patricia McLaughlin, Scott Moore, Florian Naegele, Yui Nakagawa, Falko Paetzold, Andrew Parker, Michael Peters, Manuel Rippel, Florian Rittiner, Nidhi Santen, Malte Schneider, Claude Siegenthaler, Afreen Siddiqi, Annegret Stephan, Stephan Stollenwerk, Kavita Surana, Masahito Tanada, Noriko Takemura, Tomohiro Takeuchi, Xiao Tan, Tian Tang, Margaret Taylor, Simge Tuna, Karin Vander Schaaf, Jonas Volland, Geng Wu, Xiaoqi Xu, Aya Yamamoto, Masaru Yarime and Liu Zhu. This work has benefited substantially from your time and expertise.
Monica Heinz, for being always helpful, kind and available (and for patiently forwarding me all my mail in the last year).
Jochen Markard and J. Peter Murmann, for their critical feedback at the right time.
Suzanne Greene, for carefully proof-reading several of the chapters.
The students that I had the honor to supervise, Emre Akyol, Andreas Busa, Etienne Eigle, Christian Niebuhr, Jan Ossenbrink, Fabien Richard, Malte Schierwater and Benjamin Staubli. Collectively, I probably benefited more from you than the other way around.
Above all, I want to thank my fiancée, my friends and my family for keeping me sane and for providing me with the support I needed to keep going during the highs and the lows of the PhD.
What a journey!
xi
Abstract
The urgency of the world’s environmental challenges has led governments around the globe to rethink the role of government in the innovation process. In addition to direct public spending on energy technology research, development and demonstration, many countries now subsidize the large-scale deployment of innovative energy technologies through so-called ‘deployment policies’. Examples include feed-in tariffs for renewable electricity, mandates for the blending of biofuels and investment subsidies for electric vehicles. Deployment policies can be expensive ways to mitigate the current environmental footprint of the energy system, but proponents justify them as ‘learning investments’ which will pay off in the long term as they stimulate innovation, bring down cost and enhance the performance of clean energy technologies for future generations. The debate on the validity of this claim is controversial and politicized, in part because the discussion often lumps together very different technologies and paints a simplified picture of the complex process of system transformation. This thesis presents six essays to advance this debate, three on the nature of the innovation process in different energy technologies and three on the design and governance of deployment policy instruments. Collectively, these essays make three distinct contributions.
First, this thesis introduces a novel methodology to study the evolution of technology. A combination of patent content analysis and citation-network analysis, the method developed in this thesis allows to quantitatively study the focus of research activity in a sector over time and across geographies. In the future, this methodology will allow to study a number of under-researched phenomena in the evolution of technology, including (i) the relationship between the focus of research activity and competitive advantage; (ii) the division of labor in research and development between countries and regions in global value chains; (iii) the impact of public policy on shifts in the focus of research activities.
Second, this thesis introduces the explicit consideration of differences in the innovation process between energy technologies into the analysis of deployment policies. It shows that the model of the technology life-cycle that is implicitly assumed in much of the current debate on deployment policies applies to mass-produced energy technologies, but does not adequately describe innovation in complex infrastructure technologies in the energy sector. This is important because different models of the life- cycle imply different roles for deployment – and thus deployment policies – in the evolution of technology. It means that the conceptual underpinnings of the debate on deployment policies do not apply to a significant share of the energy technology space, and calls for the explicit consideration of
xii Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact technological characteristics in decisions on deployment policy support for energy technologies, in particular the size, duration and geographical scope of support. It also allows us to reconcile conflicting evidence on the impact of deployment policies in the literature. These findings are based on three case studies in the first three essays of the thesis: (1) an analysis of the focus of inventive activity over time in wind turbine technology in 1973-2009, using a novel methodology that integrates expert assessment of patent data with patent-citation network analysis; (2) a comparison of technology life-cycles in wind turbine technology and solar PV technology in 1970-2009, using the same methodology; and (3) a techno-economic model of the impact of local and global learning on the cost of renewable energy deployment in Thailand in 2013-2021.
Third, this thesis’ results emphasize the need to consider the complex political dynamics of socio- technical transformations in the debate on energy innovation policies in general and deployment policies in particular. In most technology policy analyses, policy decisions are seen as essentially exogenous to the technological change the intervention aims to induce. In practice, this means that the possibility of changes to public policy in response to induced technological change is not part of the analysis, nor is the ability of affected actors to foresee or respond to such changes. Essay 4 demonstrates that this may not adequately reflect political reality. The essay presents a qualitative analysis of the evolution of Germany’s public policy support for solar power in 2000-2012. The essay demonstrates that deployment policy instruments can become at least in part endogenous to the transformation they were designed to induce, because they trigger unforeseen changes in the socio- technical system, and develops a model to account for these dynamics. The findings have two important implications. First, because investors in innovative energy technologies are aware of the possibility of policy changes, the effect of deployment policies on innovation will depend, at least in part, on the political system and its ability to learn and respond. Second, the ability of a policy to induce desired technological outcomes will depend as much on the design of the policy itself as on the state of the socio-technical system. Policy designs and lessons learned can therefore not always be transferred between jurisdictions – e.g., the fact that Germany may no longer need costly feed-in tariffs to attract investment in photovoltaics does not mean this policy is not the most cost-efficient option to attract investment in other jurisdictions. The last two essays explore these two implications in more detail. Essay 5 reviews and analyzes the proposals for internationally supported feed-in tariffs for renewable energy in developing countries, and discusses how to minimize the risk premium demanded by investors due to the policy risk induced by the prospect of unforeseen cost developments.
xiii
Essay 6 explores the implications of the German experience with deployment policies for solar power in the context of the newly introduced feed-in tariff for renewable electricity in Japan.
xiv Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
Zusammenfassung
In Anbetracht dringlicher globaler Umweltprobleme, wie etwa des Klimawandels und der urbanen Luftverschmutzung, wird die Rolle des Staates der Entwicklung innovativer Technologien seit einiger Zeit neu diskutiert. Neben der traditionellen öffentlichen Förderung von Forschung, Entwicklung und Demonstrationsanlagen subventionieren viele Regierungen neuerdings auch den großtechnischen Einsatz innovativer Energietechnologien durch sogenannte „Nachfrageinstrumente“, wie zum Beispiel Einspeisetarife für Strom aus erneuerbaren Energien, Mandate für die Beimischung von Biokraftstoffen oder Investitionszuschüsse für Elektroautos. Geht es nur um die Reduktion des momentanen ökologischen Fußabdrucks des Energiesystems, sind diese Nachfrageinstrumente oft vergleichsweise kostenintensiv. Befürworter argumentieren jedoch, dass es sich bei Nachfrageinstrumenten um "Lerninvestitionen" handelt, welche sich auf lange Sicht durch induzierte Innovationen, Kostenreduktionen und Effizienzsteigerungen auszahlen werden.
Die wissenschaftliche und öffentliche Debatte über den Innovationseffekt von Nachfrageinstrumenten ist kontrovers und politisiert. Dies liegt unter anderem daran, dass unterschiedliche Technologien in der Debatte nicht ausreichend differenziert werden. Auch wird oft ein sehr vereinfachtes Bild der komplexen, systemischen Transformationsprozesse und ihrer Determinanten gezeichnet. Die vorliegende Dissertation umfasst sechs Artikel, die darauf abzielen, bezüglich dieser zwei Aspekte mehr Substanz in die Debatte zu bringen. Die ersten drei Aufsätze befassen sich demnach mit technologiespezifischen Faktoren im Innovationsprozess in unterschiedlichen Energietechnologien. Die nächsten drei Aufsätze befassen sich mit der Gestaltung und zeitlichen Dynamik von Nachfrageinstrumenten im Kontext von komplexen, systemischen Transformationen und der daraus resultierenden Unsicherheit. Im Folgenden sollen die drei wichtigsten Beiträge dieser Arbeit herausgearbeitet werden.
Erstens, diese Dissertation entwickelt eine neue Methodik, mit der die Evolution von Technologien untersucht werden kann. Die Methodik integriert Ansätze aus der Patentanalyse und der Netzwerkanalyse und ermöglicht es, die Entwicklung eines komplexen technischen Systems über Zeit und in ihrer räumlichen Ausprägung quantitativ zu analysieren und graphisch darzustellen. Diese Methodik wird es in der Zukunft erlauben, eine Reihe von bisher kaum untersuchten Phänomenen genauer zu untersuchen, wie etwa (i) die Beziehung zwischen dem Fokus industrieller Forschung und Wettbewerbsvorteilen; (ii) die räumliche und betriebliche Arbeitsteilung in der Entwicklung
xv
komplexer technischer Systeme; und (iii) die Auswirkung von regulatorischen Interventionen und öffentlicher Förderung auf den technologischen Fokus der industriellen Forschung.
Zweitens, diese Dissertation stellt dar, wie relevante Charakteristika von Energietechnologien in die Gestaltung und Evaluation von Nachfrageinstrumenten mit einbezogen werden können. In den ersten drei Artikeln dieser Arbeit wird empirisch aufgezeigt, dass sich Innovationsprozesse in verschiedenen Energietechnologien deutlich hinsichtlich ihrer zeitlichen und räumlichen Muster unterscheiden. Die so herausgearbeiteten Technologiecharakteristika ermöglichen es, widersprüchliche Aussagen über die Innovationseffekte von Nachfrageinstrumenten in der wissenschaftlichen Literatur in Einklang zu bringen. Darüber hinaus ergeben sich aus einer Differenzierung von Technologien nach den herausgearbeiteten Charakteristika wichtige technologiespezifische Implikationen für die räumliche und strategische Ausgestaltung von Nachfrageinstrumenten. Die Ergebnisse beruhen auf einer Analyse des Fokus der Forschungstätigkeit in der Windturbinentechnologie im Zeitraum 1973-2009 (Aufsatz 1); einem Vergleich der Technologielebenszyklen in der Windturbinentechnologie und der Photovoltaik im Zeitraum 1970-2009 (Aufsatz 2); und einer Modellierung der Auswirkung von lokalen und globalen Lerneffekten auf die Kosten von erneuerbaren Energien in Thailand im Zeitraum 2013-2021 (Aufsatz 3).
Drittens, diese Dissertation arbeitet heraus, wie die komplexe Dynamik von systemischen Transformationsprozessen und die daraus resultierende Unsicherheit in der Gestaltung von Nachfrageinstrumenten berücksichtigt werden können. In der Innovationspolitikforschung wird die Gestaltung und der Umfang von öffentlichen Technologiesubventionen üblicherweise als exogen angenommen. Aufsatz 4 zeigt jedoch, mithilfe einer qualitativen Analyse der Entwicklung der politischen Förderung für die Photovoltaik im Zeitraum 2000-2012 in Deutschland, dass es eine starke „Rückkopplung“ zwischen Innovationpolitik und technologischem Wandel gibt. Technologische Veränderungen in der Photovoltaikindustrie in Deutschland, welche von staatlichen Subventionen ermöglicht wurden, führten wiederholt zu politischem Druck die Subventionspolitik anzupassen und zu reformieren. Das beschriebene, komplexe Zusammenspiel zwischen Innovationspolitik und technischem Wandel hat zwei Implikationen. Zum einen können Förderinstrumente nicht ohne weiteres zwischen Ländern übertragen werden, da die momentan „optimale“ Förderpolitik vom technologischen Stand der geförderten lokalen Industrie abhängt. Zum anderen hat das Zusammenspiel Auswirkungen auf die Fähigkeit von politischen Förderinstrumenten, langfristige Investitionen anzuziehen: die Erwartungen von Investoren bezüglich der politischen Nachhaltigkeit eines Subventionsregimes beeinflussen direkt die verlangte Risikoprämie in den xvi Creating Markets for Energy Innovations – Case Studies on Policy Design and Impact
Kapitalkosten, und damit die Attraktivität von risikoreichen und langfristigen Investitionen. Die letzten beiden Essays vertiefen die Diskussion der Implikationen der Aufsätze 1-4 für die Gestaltung von Nachfrageinstrumenten. Aufsatz 5 diskutiert und analysiert unterschiedliche Vorschläge für finanziell unterstützte Einspeisetarife für erneuerbare Energien in Entwicklungsländern. Der Aufsatz diskutiert wie diese Einspeisetarife gestaltet werden können, so dass die Risikoprämie, welche von Investoren verlangt wird um das Risiko nachträglicher Politikänderungen abzusichern, minimiert werden kann. Aufsatz 6 diskutiert die internationale Übertragbarkeit von politischen Instrumenten, insbesondere wie die deutschen Erfahrungen mit Einspeisevergütungen für Erneuerbare Energien auf Japans neue Förderpolitik übertragen werden können.
xvii
Synopsis
1. Introduction
1.1. Mitigating Climate Change: A Mammoth Technological Challenge
At the Earth Summit in Rio de Janeiro in 1992, the world’s leaders formalized their intention to “stabilize greenhouse gas (GHG) concentrations in the atmosphere at a level that would prevent dangerous anthropogenic interference with the climate system” as part of the United Nations Framework Convention on Climate Change (UNFCCC, 1992). At the 2009 UNFCCC conference in Copenhagen, “dangerous anthropogenic interference” was clarified when leaders agreed on the objective to keep global warming within 2°C (UNFCCC, 2009). But total anthropogenic GHGs have continued to increase since the first negotiation of the UNFCCC, as well as following Copenhagen in 2009, despite a growing number of national and international climate change mitigation policies. In fact, GHG emission growth accelerated to 2.2% in the period 2000-2010 compared with 1.3% in 1970-
2000. Roughly three quarters of this increase come from carbon dioxide (CO2) emissions from fossil fuel combustion and industrial processes, two sources which together account for about 65% of global GHG emissions (IPCC, 2014).
The time window to reduce GHG emissions is rapidly closing (Sanford et al., 2014). The Intergovernmental Panel on Climate Change (IPCC) estimates that total cumulative emissions must not exceed one trillion tons of carbon equivalents (1,000 Gt C) in order to stay, at 66% probability, within 2°C of global warming (IPCC, 2013).1 This is equal to about 30 years at current emissions levels – not much more than the 22 years the world has already spent trying to negotiate a comprehensive international climate agreement (see Figure 1). The United Nations Environment Program (UNEP) expects the world to significantly ‘overshoot’ this budget (UNEP, 2014).2
Recent international political developments seem to corroborate UNEP’s expectation. Even when taking into account the pledges made in the recent joint announcement made by China and the
1 Sustained GHG emissions at current levels are not compatible with scenarios that keep global warming within 2°C. Because a significant share of emitted GHGs remain in the atmosphere for millennia, it is useful to think in budgets for cumulative GHG emissions.
2 Under business as usual assumptions, GHG emissions are estimated to increase to 87 Gt CO2eq in 2050, compared to about 54 Gt in 2012 (UNEP, 2014).
2 Synopsis
United States, China’s GHG emissions, which now stand at 28% of the world’s total (Friedlingstein et al., 2014), will continue to rise until 2030 (The White House, 2014). Reductions from the US and Europe equivalent to China’s projected emissions growth are unlikely in that timeframe (UNFCCC, 2014; Zhang et al., 2014). Given this trajectory, the global economy will have tto become net carbon negative before the end of the century (as shhown in Figure 1) in order to stay within 2°C of global warming, e.g., through the large scale deployment of carbon-negative bioenergy (Tilman et al., 2006).
Figure 1: Observed and projected trends in global CO2 emissions under four IPCC scenarios (Sanford et al., 2014). Trends include both fossil fuel and industrial emissions, but not land use change emissions (e.g., deforesstation and wetland loss). Numbers on the right-hand side represent the median values of global mean surface temperature projections above pre- industrial levels in 2100 and the 66% probability range of the ensemble projections for each scenario. The 2046 budget number is determined from the allowable carbon emissions budget of 1,000 Pg C consistent with a >66% likelihood of limiting warming to less than 2°C.
At the same time, there is pressure to increase emissions even further. Globally over 1.2 billion people live in extreme poverty (The World Bank, 2013). 1.3 billion people – 18% of the world’s population – are without access to electricity, and 2.7 billion people rely on the traditional use of biomass for cooking, which causes harmful indoor air pollution (IEA, 2014a). In the abssence of a significant reduction in the carbon intensity of the worlld economy, understood here as tCO2eq per $ of economic
3
output, sustaining economic growth and prroviding access to energy services will lead to a strong increase in emissions. China’s economic transformation over the last three decades lifted 600 million people out of poverty and provided access to electricity to almost 100% of the country. But it also had an enormous climate footprint: China’s GHG emissions per capita are now 6% higher than those of the European Union, even though income per capita is only about one-fifth (Friedlingstein et al., 2014). Facilitating the same rates of growth in income and energy access in South America, South Asia and Africa while staying within the oone trillion ton budget will require nothing short of a technological revolution (Galiana and Green, 2009).
1.2. In Search for a Technological Revolution in the Energy Sector
Much of the technological revolution required to mitigate climate change will have to take place in the electricity, heat and transport sectors (collectively referred to in this thesis as the energy sector). In 2010, the energy sector was responsible for 65% of anthropogenic CO2 emissions (IEA, 2014b). As can be seen in Figure 2, which shows a sectoral split of CO2 emissions over time, mostt of the increase since the 1970s came from one sector: electricity generation. In fact, the recent accelleration in the 2000- 2010 decade described above can be attributed almost entirely to an increase in coal use for electricity production, in particular in China (IPCC, 22014). Overall, global total primary energy supply more than doubled between 1971 and 2012 (IEA, 22014b).
1 Figure 2: Sources of global CO2 emissions, 1970–2004 (only direct emissions by sector) (IPCC, 2007). Including fuelwood. 2Other domestic surface transport, non-energetic use of fuels, cement production and ventinngg/flaring of gas from oil production. 3Including aviation and marine transport.
Scenarios by the IPCC suggest that the electricity sector will play a key role iin mitigating climate change, ffor three reasons. First, under business-as-usual assumptions the emissions from the electricity
4 Synopsis sector will rise more rapidly than from any other sector of the economy. Second, in scenarios consistent with the 2°C target, decarbonization happens more rapidly in electricity generatiion than in any other sector, and the sector becomes net negative by mid-century. Third, low carbon electricity can reduce the cost of emission reductions in other parts of the energy sector, e.g., in the form of electric heating, electric vehicles, and increasing use of electricity in industry (IPCC, 2014).
Options exist to provide energy at low, zero or even net negative GHG emissions. As shown in Figure 3, the combined global technical potential to provide electricity, heat and transport from renewable energy sources exceeds current demand by several orders of magnitude. However, there are many social and political barriers that keep these technologies from being deployed, which mean that the development of better and cheaper technologies will not solve the climate challlenge (Unruh, 2000). Having tthat said, it is clear that clean energy technologies will need to improvee in terms of cost and performance in order to be adopted across the economy (Edenhofer et al., 2012).
Figure 3: Ranges of global technical potentials of RE sources in studies reviewed by the IPCC (Edenhofer et al., 2012). Biomass and solar are shown as primary energy due to their multiple uses. Note that the y-axiiss values are presented on logarithmic scale due to the wide range of assessed data.
Industrialized countries have sought to reduce dependency on fossil fuels in the ennergy sector since the oil crises in the 1970s. The deployment off non-fossil fuels has increased in absolute terms, largely nuclear and hydropower, but has not managed to outpace increasing demands for energy related to worldwide economic growth and development. In 2012, fossil fuels still accounted for 82% of the world primary energy supply, a share that has decreased by only 4% since 1972 (IEA, 2014b).
5
In the last decade, the deployment of clean energy technologies has accelerated, especially in the electricity sector. The International Energy Agency (IEA) estimates that investment in renewables- based power plants accounted for 58% of global power generation investment between 2000 and 2013 (IEA, 2014a). Most of this now went into wind and solar photovoltaic (PV) power, which accounted for 70% of investment in OECD countries since 2000. Overall investment in clean energy technology increased five-fold over 2000-2013, reaching a peak of $290 billion3 in 2011 ($2.3 trillion in total) (IEA, 2014a).4
Notably, a large share of investment into clean energy technologies is being made in developing countries. A group of 55 countries surveyed by Bloomberg New Energy Finance (BNEF) had 666GW of renewable electricity capacity installed in 2013, compared to a total of 806GW in OECD countries. On average, renewables (including large hydro) represented a larger percentage of total capacity in these nations than they do in the OECD (BNEF, 2014).
Despite these efforts, clean energy is not replacing fossil fuels at the rates necessary to mitigate dangerous climate change impacts (Loftus et al., 2014). Technological change must progress much more rapidly in the immediate future than it has in the last four decades. This will require substantial public policy support for innovations in energy (Trancik, 2014).
2. Policy Relevance
2.1. A Growing Role for Government in Energy Innovation
There are two widely accepted rationales for public policy support of energy innovation (Jaffe et al., 2005; Gallagher et al., 2012). First, social benefits exceed private benefits when utilities or other users adopt clean energy technologies because of distortions in the markets for energy goods and services. These market failures justify government investments to advance the public interest (e.g., Gillingham and Sweeney, 2010). Most obviously, the social cost of pollution from fossil fuels is rarely reflected in the price of electricity, heat or transport. Even in jurisdictions that put a price on pollution, e.g., through taxes or emissions trading schemes, the price signal tends to be much lower than the social cost of
3 All dollar values in this thesis are in USD. 4 90% of this went into power generation technologies, the rest into biofuel refineries.
6 Synopsis pollution and too uncertain to stimulate innovations (e.g., Schmidt et al., 2012; Taylor, 2012). But there are also more subtle ways in which markets for energy favor the technological status quo. Fossil fuel technologies are deeply embedded in formal and informal institutions in the energy sector – e.g., in standards, regulations and market power – as a result of a century-long co-evolution (Walker, 2000). The same issue is found within the physical infrastructure, where technological complementarities, economies of scale and network effects favor technologies that are widely in use over emerging alternatives (Unruh, 2000).
The second widely accepted rationale is that the social benefits from the development of new clean energy technologies exceed the private benefits of technology providers. Therefore, even if price signals in the markets for energy goods and services would be corrected for all market failures described above, firms would still underinvest in clean energy innovation (Gallagher et al., 2011). This is because, on one hand, investments in innovation are in general very risky, and options to insure this risk are limited. Furthermore, social benefits of innovation often accrue far into the future and outlive the time horizon of private investors. Lastly, knowledge generated in the innovation process can ‘spill over’ to other firms. Such knowledge spillovers can represent significant barriers, especially when investments are large and very visible, e.g., in the case of first-of-a-kind commercial-scale plants (Weyant, 2011; Nemet et al., 2014).
The world’s inability to respond to climate change has made these market failures particularly salient in the public discourse. Various stakeholders, therefore, urge governments to enlarge the size and scope of public policy support for innovation in energy, calling for public investments to accelerate different stages of the energy innovation process – in R&D, demonstration, niche markets or large- scale deployment of clean technologies (e.g., AEIC, 2010; Anadon et al., 2011; Gallagher, 2009; PCAST, 1997; Shellenberger et al., 2010). In response, governments around the globe are rethinking the role of the state in energy innovation (Mowery et al., 2010; Henderson and Newell, 2011; Mazzucato, 2011; Foray et al., 2012).
Whether public resources are better spent on research, development and demonstration (RD&D) or later stages of the energy innovation process is controversially debated in the academic community. Some argue that the “non-incremental innovation” that we need is “more responsive” to research funding and much less to investments in later stages of the innovation process (Nemet, 2009, p. 708). Others counter that climate change and resource depletion “cannot be simply researched away” (Yang and Oppenheimer, 2007, p. 203) and that “the achievement of a technical goal in the field of
7
alternative-energy technologies is only the beginning of a long process of llearning, incremental improvement, and monitoring of the performance of these technologies in a wide array of complex operating environments”, which must be supported through public investments in technology deployment (Mowery et al., 2010, p. 1201).
2.2. Shifting Priorities: From RD&D to Market Creation
While the academic debate on the relative merits of investments in energy RD&D and deployment is still ongoing,5 government actions lean heavily toward the latter.
Governments are spending a non-trivial amount of public resources on RD&D for energy technologies – in 2011, the world total was in the range of $20-30 bn (Kempener et al., 2012; IEA, 2014c). However, despite a series of calls forr a significant increase in energy RD&D (e.g., IEA, 2013; Nemet and Kammen, 2007; PCAST, 1997), absolute spending has increased only slightly over the last decade (see Figure 4). In terms of its share oof global GDP, it has even declined. Compared to other public spending priorities, energy RD&D spending looks particularly dismal: deffense alone consumes about 30% of the public RD&D& budget in IEA-member countries, almost ten times as much as energy. The IEA recommends at least three times the current spending to sstay within 2°C; the spending gap for clean vehicles and carbon capture & storage alone is estimated to be at least USD 30bn (IEA, 2013).
Figure 4: Public spending on energy research, developpment and demonstration (RD&D) in relation to other research areas, as a share of total public RD&D spending (IEA, 2013).
5 See, e.g.., the discussion section on technology poolicy and global warming in Research Policy 39 (8).
8 Synopsis
At the ssame time, governments around the globe are rapidly increasing theiir subsidies to create markets for the large-scale deployment of innovative energy technologies. In many cases, spending on these ‘deployment policies’ (Hoppmann et al., 2013) or ‘market formation policies’ (Gallagher, 2014), which includes subsidized electricity tariffs for renewable electricity, blending mandates for biofuels, and tax credits, now exceed RD&D support by orders of magnitude. Figure 5 shows projections by the IEA for subsidies supporting the deployment of renewable energy technologies. In 2013, subsidies to renewables were $121bn, 15% higher than inn 2012 and almost three times as mucch as in 2007.
These numbers mean that spending on deployment policies for renewable energy was four to six times as much as RD&D subsidies for all energy technologies combined (IEA, 2014aa). This ratio is even more skewed in certain technological areas and jurisdictions. For example, EU member countries spent 35-41 times more on deployment for solar PV and wind power than on R&D (Laleman and Albrecht, 2014); in Germany, the ratio was 100:1 for PV and 50:1 for wind power in 2011 (E-FI, 2014, p. 63). In the US, China and Japan, deployment policy support for wind and PV also outweighs R&D funding for renewable energy by at least one order of magnitude (e.g., by factor 20 in the case of wind power in the US) (GAO, 2013). This trend is likely to continue, as seen in Figure 5: the IEA predicts that between 2007 and 2040, a total of $5.1 trillion will be spent on deployment policies.
Figure 5: Historic and projected spending on deploymment policies for renewable energy. For powerr, spending is calculated as the difference between the cost of electricity and the wholesale price in each region, multiplied by the amount of generation for each renewable energy technology. For biofuels, spending is calculated by multiplying the consuumption by the difference between their production cost and the regional reference price of the comparable oil-based product in each region (IEA, 2014a).
9
Unlike traditional climate policies, these deployment policies are not only intended to stimulate the adoption of existing low-carbon technologies, but have the explicit objective to reduce costs and improve performance of clean technologies in the future. For example, the German feed-in tariff for solar power (a form of subsidized electricity tariff), at about $10bn per year the most expensive deployment policy in the world, was designed as "market entry assistance to allow for cost reductions, which will then facilitate the diffusion of photovoltaic through the market" (German Federal Diet, 1999). The $5-10bn per year of US production tax credits for renewable electricity was enacted to enable “further advances of renewable energy technologies” (102nd Congress, 1992). And the US tax credit under the US ‘Recovery Act’ in 2009, with a total of $7.5bn per year, had the objective “to help renewable energy technologies achieve economies of scale and bring down costs” (The White House, 2009).
Deployment policies are thus best understood as ‘learning investments’ (Sagar and van der Zwaan, 2006), rather than as static climate policy instruments. Indeed, if seen as static instruments to reduce current GHG emissions, most of current deployment policies are prohibitively expensive. At the beginning, the GHG abatement cost of the German support for solar power were as high as 760€ per ton of CO2 (Frondel et al., 2008). This is orders of magnitude higher than the CO2 emission allowance prices under the EU’s Emission Trading Scheme, which provides a proxy for marginal abatement cost in the EU and has never exceeded 50€ per ton (since the end of 2012, they are below 10€; The WorldBank, 2012).
Broadly speaking, the motivation to fund large-scale deployment in order to advance technology is rooted in two empirical observations. First, numerous empirical studies suggest that the costs of a technology tend to correlate negatively with cumulative production, as captured in so-called experience curves (e.g., Wene, 2000). Figure 6 illustrates this widely observed phenomenon for a few examples from different sectors: electromechanical equipment (coal plants), biochemistry (ethanol), semiconductors (solar cells and transistors). A report in 1999 by the United States’ President’s Committee of Advisors on Science and Technology (PCAST) summarized this motivation in a report on energy technology policy:
“Once a technology has been demonstrated at a potentially commercially viable scale, there remains a long process of building a series of such systems to scale up equipment manufacturing facilities and also to learn how to reduce manufacturing, system installation, and operations and maintenance costs to competitive levels. …
10 Synopsis
Many products, for instance, have costs that drop by 10-30 percent for every doubling of cumulative production. To move a new technology into the market, its higher initial costs relative to comppeting products must be covered. As cumulative production volume increases, costs will be reduced until some innovaative energy technologies become fully competitive with conventional technologies..” (PCAST, 1999, ES-7).
Second, globally successful manufacturers of new technologies often emerge in countries that support deployment of these technologies early on, thereby creating ‘home markets’ that give domestic firms a lead in innovation (Beise-Zee, 2004). The Danish wind turbine industry, for example, benefited from an early home market that was supported by government policies, and mainttained a large global market share even when domestic demand declined (Andersen, 2004). This and other success stories have motivated countries to pursue deployment policies in the energy sector (Matthews and Tan, 2014).
Figure 6: Empirical experience curves for different technologies (McNerney et al., 2011). Each curve was rescaled and shifted to aid comparison with a power law. Tick markks and labels on the left verticall axis show the ffirst price (in real 2000 dollars) of the corresponding time series, and those of the right vertical axis show the last price. Linees are least-squares fits to a power law.
The new focus on deployment as a means to stimulate innovation represents a stark deviation from existing technology policy strategies, and reflects that governments are rethinkingg traditional models of technology policy in the context of global warming (Mowery et al., 2010). However, the theoretical understanding of the impact these learning investments has not kept pace with theeir application. There is relatively little academic research on deployment policies (e.g., Nemet, 2009; Peters et al., 2012; Bettencourt et al., 2013; Hoppmann et al.., 2013, 2014). And those studies tthat have been done
11
present conflicting evidence (cf., Nemet, 2009; Bettencourt et al., 2013) and are scattered across academic (sub-) disciplines.
This dissertation assumes that the world is going to spend large sums of public resources on deployment policies in the energy sector, and aims to contribute to the understanding of how these policies can stimulate innovation. The objective is to facilitate better deployment policy design in the future.
3. Research Design
This dissertation addresses three overarching research questions that are rooted in the empirical challenges that policymakers face in the context of climate change. This section first puts these high- level research questions into context (3.1), and then explains the theoretical perspective and research framework adopted by this thesis (3.2 and 3.3). The specific research question and empirical strategy of each individual paper are introduced in section 3.4 and 3.5.
3.1. Overarching Research Questions
Governments plan to spent large sums of public resources on deployment policies to induce innovations in clean energy technologies, as described in section 2. In light of the magnitude of spending and the urgency of climate change, is it is important that these public resources are spent effectively. The overarching research question of this thesis is therefore:
R0: How can deployment policies be designed in order to maximize their impact on innovation?
In broader perspective, the government-led creation of markets for innovative technologies in general is not new (Edler and Georghiou, 2007). Governments have long made use of demand-side measures to stimulate innovation, e.g., through targeted public procurement, tax incentives or mandates for the adoption of certain safety or environmental technologies (see Edler, 2013 for a comprehensive overview). But the recent use of deployment policies in the energy sector stands out in two important ways: the scope and scale of innovation that these policies aim to induce. This thesis aims to explore these two aspects in detail.
12 Synopsis
3.1.1. Accounting for Technological Characteristics in Deployment Policies
Deployment policies in the energy sector are used to stimulate innovation across a very broad scope of technologies through relatively standardized policy instruments. Unlike other sectors, such as chemicals or semiconductors, the energy sector is not defined by a specific field of knowledge. A huge variety of technologies from almost all sectors of the economy are employed in the extraction, conversion and end-use of energy. The actors in the sector itself, such as electric utilities, district heat providers or transport system operators often only develop relatively few technologies themselves (Markard, 2011; Wiesenthal et al., 2011).6 Rather, most energy innovations enter the sector embodied in specialized equipment or innovative fuels from supplying industries, such as semiconductors (solar cells), electro-mechanical machinery (gas turbines), agriculture (biofuel feedstock) and biochemistry (biofuel conversion). Some technologies are mass-produced (e.g., LEDs); others are large infrastructure systems (public transport systems). Some are very specific to local geographies, such as biofuel feedstock production techniques, while others are globally applicable almost without adaptation, such as the feedstock-to-biofuel conversion in bio-refineries. Deployment policies in the energy sector thus aim to stimulate innovation not in a narrowly defined industrial context, but across a wide range of sectors characterized by different knowledge bases, value chain structures and innovation processes (Malerba, 2002).
At the same time, most deployment spending is concentrated in a relatively limited number of policy instruments, some of which are used in a one-size-fits-all fashion. For example, fixed feed-in tariffs for renewable electricity alone account for more than half of projected deployment subsidies and are typically offered to all forms of renewable electricity. This differentiates deployment policies in the energy sector from public procurement instruments in, say, defense: there, technology is also sourced from a wide range of sectors, but procurement policies are typically designed in the form of contracts tailored to the characteristic needs of individual departments (Mowery, 2012). This thesis therefore aims to address the following research question:
RQ1: How can deployment policies be designed to account for technological characteristics?
6 In Pavitt’s taxonomy, the energy sector is ‘supplier-dominated’ (Pavitt, 1984).
13
3.1.2. Ensuring Effectiveness over the Lifetime of Deployment Policies
Projections suggest that, in order to compete with fossil fuels, low-carbon technologies have to be supported over a long period of time, and eventually brought to very large scale. The projected scale of public spending on individual deployment policy instruments in the energy sector appears unprecedented, even by the standards of defense or space RD&D (see section 2.2), and creates two concrete challenges for the design and governance of deployment policies.
First, many deployment policies have a time horizon and thus aim to support innovation across many product and process generations as firms move down the experience curve, rather than targeting individual products such as; say, a new jet fighter or rocketship. Most energy products and services are commodities – e.g., fuels, electricity and heat – that allow for relatively little product differentiation. Therefore few self-sustaining niche markets exist and technologies have to be supported until they can compete on costs with fossil alternatives. The design of these policies needs to account for the fact that the technological and socio-economic risks and challenges that innovators face change over time as the supported technology evolves from a few small demonstration plants to the point where it is deployed throughout the economy (e.g., Jacobsson and Bergek, 2004). However, in most technology policy analyses, policy decisions are seen as essentially exogenous to the technological evolution that the intervention aims to induce.
Second, since investments in the sector are capital intensive and have long lifetimes, only relatively predictable policy support can have the desired impact on innovation. At the same time, as mentioned in section 2.2, deployment policies are typically expensive ways to mitigate carbon emissions in the near term, which makes the short-term cost of deployment policies very salient and controversial in the public debate. Their benefits in the form of innovation and long-term cost reductions, on the other hand, are inherently uncertain and difficult to determine, even in hindsight. For example, a recent report by the German government’s Expert Commission for Research and Innovation analyzed patenting in renewable energy technologies in Germany and came to the conclusion that “there is no measurable innovation impact” of the German support for renewable electricity (a 15-year old program that cost about $20bn per year), which caused a stir in the media (E-FI, 2014, p. 15). A large group of renowned innovation researchers promptly replied with the statement that there is, in fact, plenty of evidence when looking at a broader range of indicators (Ragwitz et al., 2014).
14 Synopsis
The challenges created by large costs and intangible benefits are aggravated by the fact that most deployment policies are designed not as one-off grants, but as long-term subsidies for which any investment that fulfills certain technological criteria is eligible. Under these circumstances, if the policy is unexpectedly effective in attracting investments, large subsidy commitments over decades can be locked-in in only a few years (Frondel et al., 2009). Such cost overruns can be very difficult to correct. Several large deployment policies have been cancelled abruptly or retroactively changed because of such unpredicted cost developments. Prominent examples include the policy support for renewable power in Spain and the Czech Republic, where promised long-term subsidies were retroactively reduced, leaving investors with stranded assets. In other cases, deployment policy support was significantly reduced for new investments, leading to a steep decline in demand – e.g., in the case of the ‘California Wind Rush’ in the early 1980s or the German support for biomass power in 2014 – leading to bankruptcies of project developers, technology suppliers and service providers. Both scenarios put the policy’s long- term impact on innovation in question and risk the perceived and/or actual waste of scarce public resources. What’s more, since the investors’ cost of capital is affected by their and the lenders’ degree of trust in the policy, even the possibility of retroactive changes could prevent the deployment policy from attaining its targets cost-effectively.
In view of the need to design deployment policies that remain effective in inducing innovation at very large scale over the entire policy lifetime, it is important to understand how deployment policies can (i) adapt to changing support needs as the technology evolves, and (ii) be designed to minimize political uncertainty as well as the risk of declining legitimacy and retroactive changes. Therefore, the second overarching research question explores the implications for public policy design of the dynamic perspective on deployment policies:
RQ2: How can deployment policies be designed to ensure effectiveness over the lifetime of the policy?
3.2. Theoretical Perspective
To translate the overarching research questions into concrete schemes of analysis for the individual case studies, this thesis applies a theoretical perspective rooted in the field of innovation studies (Fagerberg and Verspagen, 2009; Fagerberg et al., 2012; Truffer et al., 2012). The theoretical concepts primarily employed here are related to (i) the evolution of technology and (ii) the broader socio-technical system in which this evolution is embedded. These two blocks of concepts inform how RQ1 and RQ2/3 are addressed, respectively.
15
3.2.1. Technological Evolution as Learning Process
Technological products are conceptualized in this thesis as systemic artifacts (Saviotti, 1986; Tushman and Rosenkopf, 1992), consisting of a non-trivial number of interdependent sub-systems and components. These sub-systems and components are organized by a product architecture, which allocates system functions to the individual components and defines the interfaces between them (Simon, 1962; Clark, 1985; Baldwin et al., 2014).
Apart from rare periods of competition between different architectures, innovation in technological products is understood as proceeding predominantly through the refinement within and extension of existing product architectures, in a cumulative and incremental learning process (Nelson and Winter, 1977; Dosi, 1982). Shaped by an understanding of important technological bottlenecks and promising avenues of solutions widely shared within the community of practitioners, the learning process focuses on only a small fraction of a product’s components and possible directions of change at any point in time (Rosenberg, 1969; Hughes, 1992; Ethiraj, 2007; Dedehayir and Mäkinen, 2011). Other parts of the product – and, crucially, how the components work together – are retained essentially unchanged (Anderson and Tushman, 1990; Henderson and Clark, 1990; Murmann and Frenken, 2006). The resulting ‘ordered’ pattern of technical change is referred to as the technological trajectory. The process of technological learning along this trajectory involves numerous feedback loops between research, development, demonstration and deployment as firms experiment with new components and sub- systems (e.g., Kline and Rosenberg, 1986), which blurs the boundaries between invention, innovation and diffusion (Fleck, 1988).
This perspective implies that, in order to understand the role of technology deployment for innovation, it is essential to analyze the learning processes in different energy technologies including the nature of feedback loops between deployment on one side and research, development and demonstration on the other. More specifically, in order to address the question how deployment policies in the energy sector can be designed to account for technological characteristics (RQ1), this thesis analyzes how technological characteristics affect the (i) sequences in the process of learning (e.g., the early aircraft industry focused first the aerodynamic characteristics of the body, then on improving the engine); (ii) type of learning process (e.g., learning-by-doing, learning-by-using); and (iii) spatial patterns of learning (e.g., learning in local clusters of firms or learning in global communities of practitioners).
16 Synopsis
3.2.2. Dynamics in Socio-Technical Systems
Following the literature on innovation systems, technological evolution is understood here as embedded in socio-technical systems, formed by a large number of actors (e.g. firms, policymakers), networks (formal and informal), technologies (knowledge and artifacts) and institutions (e.g. norms, values or regulations) (e.g., Carlsson and Stankiewicz, 1991; Edquist et al., 2005). Accordingly, the ‘outcomes’ of technological evolution, i.e., the development, adoption and diffusion of technologies, are the result of dynamic interactions between the individual elements of the socio-technical system. These interactions are understood to be non-linear and to involve numerous externalities (positive and negative), thus giving rise to emergent system properties such as bottlenecks and virtuous cycles (Bergek et al., 2008; Negro et al., 2012).
From this perspective, deployment policies can play a crucial role in technological evolution whenever they help overcome bottlenecks and initiate virtuous cycles (Bergek et al., 2008; Wieczorek and Hekkert, 2012). However, at the same time, policy interventions and policy changes cannot be seen as entirely exogenous to the evolution of the socio-technical system (Kern, 2011; Meadowcroft, 2009, 2011; Scrase and Smith, 2009). On one hand, policy makers may hold differing opinions on what constitutes the most important bottlenecks and how to remove them (Meadowcroft, 2009). What is more, even if there is a political consensus regarding the goals and means of policy-making, the inherent complexity of socio-technical systems may limit the degree to which consequences of policy interventions can be accurately foreseen (Faber and Alkemade, 2011). On the other hand, the political legitimacy of continued policy intervention can erode over time in response to unforeseen or unpopular technological outcomes.
This thesis aims to inform the design of deployment policies that attempt to support technologies with large subsidies over decades and to bring technologies from small niche markets up to deployment throughout the economy. From the systems perspective outlined above, the degree to which deployment policies can be effective in supporting innovation over the lifetime of the policy (RQ2) depends essentially on two factors, which will be addressed in this thesis: whether they successfully address the series of bottlenecks as the socio-technical system evolves; and whether they account for existing system structures (including technologies, actors, networks and institutions) so as to maintain political legitimacy over the full lifetime of the policy intervention.
17
3.3. Contributions to the Literature
Each individual chapter makes a specific contribution to the literature on technological innovation and public policy. That said, there are two overarching contributions from the thesis as a whole to the literature on technology policy: (i) the explicit consideration of linkages between technological characteristics, learning processes and deployment policy impacts (related to RQ1), and (ii) the explicit consideration and anticipation of how technological change feeds back into the political process and affects policy decisions (RQ2).
The two contributions are illustrated in two figures below. Figure 7 shows the literature context of the papers addressing RQ1 as well as the contribution of this thesis. The figure illustrates that there are a number of studies that analyze the impact of deployment policies on technological invention and innovation (e.g., Nemet, 2009; Johnstone et al., 2010; Peters et al., 2012; Hoppmann et al., 2013). There are also several in-depth studies of learning processes in energy technologies (e.g., Wilson, 2012) and the relationship between learning processes and the impacts of technology policy in general, and deployment policies in particular (e.g., Norberg-Bohm, 2000; Winskel et al., 2014). However, there are only few studies that analyze how technology-level characteristics, such as the product architecture and the cost structure, affect learning processes and deployment policy impacts, which is the focus of the papers addressing RQ1.
Existing analyses of policy impact on technological change (e.g., Hoppmann et al. 2013; Johnstone et al. 2010; Nemet 2009; Peters et al. 2012)
Deployment policy Technological change
Contribution of this thesis (RQ1) Existing analyses of technology-specific patterns in Learning process technological learning and their Technological characteristics implications for policy impacts characteristics (e.g., Winskel et al. 2014; • Learning sequence • Product Norberg-Bohm 2000) • Type of learning architecture process • Cost structure • Spatial patterns
Existing analyses linking technology characteristics and learning processes (e.g., Wilson 2012)
Figure 7: Unique contribution of papers addressing RQ1 to the literature and differentiation from existing analyses.
Figure 8 shows the literature context of the papers addressing RQ2 and the gap that this thesis addresses. Building broadly on the work of Lindblom on ‘muddling through’ (Lindblom, 1959), there
18 Synopsis have been a number of studies in the political science literature of the dynamics of the policymaking process and learning on the side of policymakers (Bennett and Howlett, 1992; May, 1992; Sanderson, 2002). There have also been studies of the impact of deployment policies on broader socio-economic dynamics of technological systems, including the formation of actor coalitions and political legitimacy (Jacobsson and Bergek, 2004; Jacobsson et al., 2004), and the feedback of those impacts to the political process (Jacobsson and Lauber, 2005, 2006). However, there have been few studies that looked at the impact of technological change on the dynamics of policymaking in technology policy in general and deployment policies in particular.
Contribution of this thesis (RQ2)
Existing analyses of systemic impacts of deployment policies Technological change (e.g., Jacobsson et al 2004; Jacobsson & Bergek 2004)
Deployment policy Transformation of socio-technical system
Actors, networks & Literature on policy learning (e.g., Bennett & institutions Howlett 1992; May 1992; Sanderson 2002).
Existing analyses of systemic impacts of deployment policies (e.g., Jacobsson & Lauber 2005, 2006)
Figure 8: Unique contribution of papers addressing RQ2 to the literature and differentiation from existing analyses.
3.4. Research Framework and Contributions of Individual Papers
Figure 9 presents the research framework of this dissertation, indicating the main concepts and relationships. It merges Figure 7 and Figure 8, the two figures representing the contributions of RQ1 and RQ2. The focus of the individual papers is indicated by the numbers in brackets.
In the big picture, this thesis is concerned with the impact of deployment policies on technologies change, understood here as encompassing invention, innovation and diffusion. More specifically, this thesis aims to inform deployment policy design in view of technological characteristics (shown in the lower part of Figure 9) and the interplay between technological change and the policymaking process (upper part of Figure 9).
19
Papers (4, 5, 6)
RQ2
Deployment policy Technological change
RQ1 (2, 3)
Technological Learning Process characteristics Characteristics
• Product architecture • Learning sequence • Scale of production (1,2) • Type of learning process • Cost structure • Spatial patterns
Figure 9: Research framework of this thesis, indicating main constructs and relationships. “RQ1” and “RQ2” mark the mechanisms analyzed to address the two overarching research questions. The focus of the individual papers is market by the numbers in brackets.
The role of each of the six papers in the framework is summarized in Table 1. While there are a number of linkages and overlaps between the individual papers, the thesis is best understood as divided into two sets of three papers, where each set addresses one of the overarching research questions.
Papers 1-3 investigate the linkages between technological characteristics (in particular, the product architecture and cost structure of a technology), prevalent learning processes (sequences in learning, types of learning processes and spatial patterns of learning) and the role of deployment for technological change (RQ1). Paper 1 analyzes the evolution of wind power technology to draw conclusions about how technological characteristics affect the shape of the technological trajectory and the aggregate, industry-wide learning process. Paper 2 builds on the findings and methodology of paper 1 and compares the learning processes in two technologies (wind power and solar PV), drawing conclusions about the influence of technological characteristics on the role of deployment for innovation in different energy technologies. Paper 3 models the experience curve of six renewable energy technologies in a developing country in order to shed light on the impact of technological characteristics on (i) the importance of local and global learning and (ii) the role of local and global deployment for innovation.
Papers 4-6 are primarily concerned with the interplay over time between deployment policies and technological change, and the implications of such a dynamic perspective for deployment policy design (RQ2). Paper 4 maps the co-evolution of the German feed-in tariff policy for solar power and the supported socio-technical system over time and identifies feedbacks from induced technological
20 Synopsis change to the policymaking process. Paper 5 models the cost drivers of renewable energy deployment policies in developing countries and discusses how the cost uncertainty can be anticipated and managed in order to enhance the effectiveness of such policies. Finally, paper 6 explores the lessons learned from the German feed-in tariff policy for solar power (paper 4) for the design of the new Japanese feed-in tariff policy.
Table 1: Overview of the six papers. Contribution of individual papers to overarching research questions is marked in column “RQ”.
# Title RQ Role in Thesis / Framework Research Case Methodology
How a Product’s Design Hierarchy Patent-citation network Shapes the Evolution of Develops a methodology to analyze the impact of Wind power analysis, 1 Technological Knowledge – 1 technological characteristics on the learning process in (global) Qualitative patent Evidence from Wind Turbine energy technologies and applies it to one specific case content analysis Technology
Applies the methodology developed in paper 1 to Technology Life-Cycles in the Patent-citation network compare two energy technologies, and analyzes the Energy Sector – Technological Wind power and analysis, 2 1 impact of technological characteristics on (i) the Characteristics and the Role of solar PV (global) Qualitative content learning processes in different energy technologies and Deployment for Innovation analysis (ii) the role of deployment policies for innovation
The Effect of Local and Global Six renewable Analyzes the impact technological characteristics on (i) Learning on the Cost of energy 3 1 the importance of local and global learning and (ii) the Techno-economic model Renewable Energy in Developing technologies in role of local and global deployment for innovation Countries Thailand
Compulsive Policy-Making – The Qualitative analysis of Analyzes the interplay between deployment policies and Evolution of the German Feed-in Solar PV in interview data, 4 2 technological change, especially feedbacks from induced Tariff System for Solar Germany parliament debates and technological change to the policy process Photovoltaic Power secondary sources
Builds on findings from papers 3 and 4 and International Support for Feed-in Six renewable demonstrates that the effectiveness of deployment Tariffs in Developing Countries – energy 5 2 policies in developing countries depends on uncertain Techno-economic model A Review and Analysis of technologies in future cost developments, and discusses how this finding Proposed Mechanisms Thailand can be translated into policy design
Builds on the lessons learned in paper 4 and discusses Japan’s Post-Fukushima how the effectiveness of deployment policies depends on Challenge–Implications From the 6 2 Legitimacy stemming from domestic industry creation, Solar PV in Japan Comparative analysis German Experience on Renewable and how this relationship can be managed in the case of Energy Policy Japan
21
3.5. Methodology
This thesis investigates the complex interplay between public policy and technological change from multiple perspectives. Given the complementarity of quantitative and qualitative methods, a combination of both is fruitful (Creswell, 2013) and can yield better results than an analysis based on a single method (Taylor et al., 2005).
Overall this dissertation employs four methodologies (see Table 1 above). Papers 1 and 2 integrate quantitative citation-network analysis with qualitative content analysis of patents into a novel methodology to study the evolution of technology. This methodology allows a quantitative analysis of the technology life-cycle and allows one to draw comparisons between energy technologies about importance of different learning mechanisms over time and the role of deployment in innovation. Papers 3 and 5 are based on a bottom-up techno-economic model of Thailand’s Alternative Energy Development Plan, a deployment policy for six renewable energy technologies in the country’s electricity sector. The model allows for the development empirically grounded estimates of the effects of various factors, including technological characteristics and different learning mechanisms, on the cost of achieving the targets.
The remaining two papers (4 & 6) employ qualitative methodologies. Paper 4 is based on a qualitative case study using interview and archival data (especially parliamentary debates) on Germany’s solar power policy. Paper 6 explores some of the lessons learned in the work for Paper 4 in a comparative study of the socio-economic context for solar power in Germany and Japan.
4. Summary of the Papers’ Findings
This section provides a brief summary of each paper’s main findings, and explains how they build upon and relate to each other. Each paper’s summary contains the implications that are relevant for the understanding of the other papers’ findings. Broader policy implications that address the overarching research questions RQ1 and RQ2 are discussed in section 5.
22 Synopsis
4.1. Paper 1: How a Product’s Design Hierarchy Shapes the Evolution of
Technological Knowledge – Evidence from Wind Turbine Technology
Paper 1 analyzes the technological trajectory in wind turbine technology between 1973-2009 to shed light on the relationship between technological characteristics and the long-run dynamics of technological learning in energy technologies (Huenteler et al., 2014b). To do so, a novel methodology is developed that integrates tools from (i) network analysis and (ii) patent-content analysis. This method allows us to study the technological focus of inventions along the core trajectory of technological change in an industry. The paper’s main findings center around the temporal pattern of innovation along a technological trajectory and its main determinants. In particular, the results demonstrate that that the sequence of technological problems solved in an industry is not random, but a function the product architecture and the demanded functional characteristics – i.e., a combination of supply-side and demand-side factors.
As an illustration, Figure 10 below shows inventive activity in different sub-systems of a wind turbine over time. It can be seen how the focus of inventive activity shifted from the rotor, which directly affects most functional characteristics (i.e., the ‘core’ sub-system), over time to more peripheral sub- systems (first power-train, then grid connection and mounting & encapsulation). Overall, it took a little over 20 years from the onset of large-scale deployment of wind turbines until the point when the focus of inventive activity had shifted through all sub-systems of the product architecture.
Focus of patenting in wind power over time
% of patents (weighted) 100%
80% Rotor 60% Power train Mounting & encapsulation 40% Grid connection
20%
0% 1975 1980 1985 1990 1995 2000 2005 2009
Figure 10: Focus of main patents in wind turbine technology, illustrating the shifting focus of inventive activity over time.
These findings have two important implications for the analysis of deployment policies. First, technological characteristics shape the temporal patterns of technological evolution. The design of
23
energy technologies with complex product architectures, such as wind turbines, continues to evolve decades after the onset of large-scale deployment, even if a dominant design has been reached for core sub-systems (such as the three-bladed rotor in the case of wind turbines). If designed appropriately, deployment of these technologies can subsequently unlock technological potentials by enabling continuous experimentation with novel sub-system and component designs, rather than simply induce cost reductions. This poses interesting questions for deployment policy design. In particular, how can large-scale subsidy schemes be designed to stimulate experimentation, encourage user-producer interaction and reduce technological uncertainty? Section 5.1 below discusses this question in detail.
Second, the demanded performance characteristics have a direct influence on the direction of technological evolution. This suggests that the nature of demand created through deployment policies matters for their impact on technological change at least as much as the size of demand (e.g., whether investors are incentivized to demand grid-friendly electricity or low cost). As described in section 2.2 above, most deployment policies aim to induce cost reductions by subsidizing learning and economies of scale because technological progress is seemingly mostly about cost reductions. For example, financial incentives for wind turbines in many countries are designed in a way that investors do not have to worry about the intermittency of wind power or the grid behavior more generally. These deployment policies miss out on opportunities to stimulate technological evolution in directions that might be beneficial in the long term, at least in the case of complex products such as wind turbines.
4.2. Paper 2: Technology Life-Cycles in the Energy Sector – Technological
Characteristics and the Role of Deployment for Innovation
The second paper builds on the findings and the methodology from Paper 1 and presents a comparative analysis of the technological trajectories in wind power and solar PV (Huenteler et al., 2014c). The findings are related to the temporal patterns of learning as well as the predominant type of learning process (learning-by-doing, learning-by-using) along the technological trajectories.
In particular, the paper analyzes which of two common models of innovation over the technology life- cycle best describes the pattern of innovation in the two technologies. The results suggest that solar PV technology followed the life-cycle pattern of mass-produced goods, a model that typically applies to technologies with relatively simple product architecture and a large-scale production process: early product innovations were followed by a surge of process innovations, especially in solar cell production. Wind turbine technology, in contrast, more closely resembled the life-cycle of complex products and
24 Synopsis systems, a model that has been developed for technologies with a complex product architecture and low-volume production: the focus of innovative activity shifted over time through different parts of the product, rather than from product to process innovations.
The results are based on an analysis of the ‘critical path’ of the patent-citation networks (a proxy for the main trajectory of technological development) in the two technologies over the period 1963-2009. Figure 11 below shows how the critical paths of the two technologies shift over time from product to process focus (solar PV, on the left) and through different parts of the system (wind power, on the right).
a) Solar PV b) Wind power PV system Wind turbine architecture architecture
Cell Rotor
Module Power train
Mounting system Grid connection
Mounting & Grid connection encapsulation
Sequence of patents along critical path Sequence of patents along critical path
Patent relating to product innovation (size ~ node weight) Citation ≤ 5 year lag (width ~ link weight) Legend: Patent relating to proces innovation Citation > 5 year lag
Figure 11: Focus of innovative activity in solar PV and wind power along the technological trajectory.
The findings have implications for the patterns of technological learning in energy technologies from the general literature on technology life-cycles. In solar PV, most innovations after the first large-scale deployment of the technology in the 1980s were focused on the production process, which points to a predominant role of learning-by-doing, economies of scale and innovations in production equipment. In wind power most innovations introduced novel sub-system and component designs, which highlights the importance of learning-by-using, technology up-scaling and innovations in operation & maintenance (O&M). These differing patterns correspond well with existing studies of technological learning in the two technologies and help to put these studies in comparative context.
The contrasting characterizations of the learning processes in the two technologies suggest that deployment policies play very different roles in innovation in the two technologies: in a learning process that is centered on the production process, deployment policy support can be crucial to enable learning-by-doing, large-scale production and markets for production equipment. In contrast; in a learning process that is centered on the product design, deployment policy support can be crucial to
25
enable learning-by-using, gradual up-scaling and markets for specialized O&M service providers. What these differing roles of deployment policy support mean for the design of policy instruments is discussed in section 5.1 below.
4.3. Paper 3: The Effect of Local and Global Learning on the Cost of Renewable
Energy in Developing Countries
The third paper focuses on the relationship between technological characteristics and the spatial patterns of technological learning of different energy technologies in advanced developing countries (Huenteler et al., 2014a). The paper presents a quantitative case study of Thailand Renewable and Alternative Energy Development Plan (AEDP; 2013-2021). In particular, it develops a model of Thailand’s electricity sector to estimate the relative contributions of local and global learning to cost reductions in six renewable electricity sources supported under the AEDP and analyzes the differences between the technologies.
Figure 12 illustrates the three main findings of the paper. The different bars show projections for the cost of electricity in 2021, in c$ per kWh, for 6 renewable and 8 fossil generation technologies. For each renewable energy technology, four bars depict the results of four different model specifications: one without any technological learning over the lifetime of the policy (2013-2021), one with only local learning, one with only global learning and one with both.
First, technological learning can, in the near future, reduce the cost of renewable electricity in emerging economies to a level that is close to competitiveness with fossil fuels. Second, in aggregate, the largest potential for cost reduction lies in local learning. This finding lends quantitative support to the argument that the conditions enabling local learning, such as a skilled workforce, a stable regulatory framework, and the establishment of sustainable business models, have a more significant impact on cost of renewable energy in developing countries than global technology learning curves. The recent shift of international support under the UNFCCC toward the strengthening of local innovation systems is therefore promising. Third, relative importance of local and global learning differs significantly between technologies. The results show that local learning is most important, in relative terms, for micro hydro, biogas and biomass. Global learning is most important for solar PV, concentrating solar and wind power.
26 Synopsis
Levelized cost of renewable electricity in 2021 in different models
LCOE for [in cUSD2012 per kWh] No learning Local learning Global learning Local and global learning
35
30
25 Marginal cost of 20 displaced electricity* 15
10
5
0 Wind Solar PV CSP Biomass Biogas Micro hydro IGCC Nuclear CCGT gas Gas turbine Gas IGCC lignite IGCC Subcritical coal Subcritical Supercritical coal Supercritical Subcritical lignite
Figure 12: Impact of local and global learning on the cost of renewable electricity technologies under Thailand’s Alternative Energy Development Plan. The four different bars for the six renewable energy technologies represent different models.
The results for the relative impact of local and global learning depend on assumptions about how much of the technology is manufactured locally. To obtain realistic estimates for the spatial patterns of learning, the paper compiled typical cost splits between locally and globally sourced components for the six technologies in an advanced developing country like Thailand (shown in Table 1). The estimates suggest that the relative contributions of local and global learning are affected by technological characteristics, in particular the product architecture and the cost structure, which affect the share of components and services that can typically be manufactured locally: All components and services that require a high share of localized knowledge (say, about local geography or regulations), are produced in very small volumes or costly to transport are typically sourced locally. Only those components and services that are relatively standardized and large-scale in production are sourced globally.
27
Table 2: Split between locally and globally sourced components for the six analyzed technologies.
Cost split Learning rate Technology Locally sourced parts Globally sourced parts (local/global) (local/global)
Nacelle (including electrical Grid connection; engineering; procurement & Wind machinery, power electronics & 67%/33% 11.3/4.3 construction; foundation; rotor blades; tower control system)
Grid connection; engineering; procurement & PV PV modules; inverter 43%/57% 17/20 construction; balance of system excluding inverter
CSP Grid connection; engineering; procurement & Power block; heat transfer fluid cycle 67%/33% 14.6/14.6 (solar tower) Construction; solar field
Biomass Grid connection; engineering, procurement & Steam turbine and electric generator (anaerobic construction; fuel shredder; boiler; heat exchanger; (prime mover); flue gas and water 75%/25% 5/5 digestion) piping treatment
Grid connection; engineering, procurement & Gas engine (prime mover); 25% of Biogas construction; fuel handling; balance of system; 75% of 78%/22% 5/5 converter system converter system
Grid connection; engineering, procurement & Micro hydro 50% of electro-mechanical equipment 87%/13% 5/5 construction; 50% of electro-mechanical equipment
For sources, see Paper 3.
Similar to papers 1 and 2, this paper’s findings suggest that optimal deployment policy design can look very different for different energy technologies. The characteristics that determine local and global sourcing are very similar to the characteristics that affect the technology life-cycle pattern, as discussed in Paper 2. In general, mass-produced goods, with low architectural complexity and a high scale of production, such as solar PV, fuel cells or LEDs, will have a higher share of globally sourced components and services and benefit more from global learning effects. Complex products and systems, on the other hand, with high architectural complexity and low scale of production, tend to benefit more from local and regional learning effects. How these spatial patterns of learning translate into implications for deployment policy design is discussed in section 5.1 below.
4.4. Paper 4: Compulsive Policy-Making – The Evolution of the German Feed-in
Tariff System for Solar Photovoltaic Power
Paper 4 analyzes the interplay between policy design and technological change to shed light on the long-term dynamics of deployment policy interventions (Hoppmann et al., 2014). It presents a case study of the German feed-in tariff system for solar power in the period 2000-2012, a highly effective and widely copied policy instrument targeted at fostering the diffusion and development of renewable
28 Synopsis energy technologies. The analysis demonstrates that the policy has been subject to a considerable amount of changes, many of which are the result of policy makers addressing specific system issues and bottlenecks. Interestingly, however, often these issues themselves were driven by unforeseen technological developments induced by previous policy interventions.
A key motivation for enacting the German feed-in tariff for solar power in 2000 was to stimulate technological change in PV technology by enabling firms to enter the stage of mass production. However, over the 12 years that followed, the technology evolved in several ways that were not foreseen by the designers of the policy. First, the number of PV installations increased much faster than expected, culminating in debates about the costs connected to technology support that had to be borne by electricity consumers. Second, costs of PV technology fell at a considerably higher rate than predicted and resulted in windfall profits for both producers and users of PV technology. Third, especially in recent years, increasing deployment of PV has raised concerns about the stability of distribution grids and the longer-term integration of renewable sources into the electricity market. Fourth, while initially the German industry performed well, in recent years, a strong Chinese industry has emerged that markets its products in Germany, thereby directly profiting from the demand-side subsidies put in place.
The German government responded to each of these issues by implementing legislative or administrative changes to the feed-in tariff policy. These responses were often successful in resolving the immediate bottleneck, e.g., reducing windfall profits, and have been crucial in sustaining the effectiveness and political legitimacy of the German feed-in tariff policy, until it was essentially abolished in 2014. But they proved inadequate to mitigate foreseeable but less urgent issues. In fact, the analysis shows that, in many cases, policy interventions often triggered changes in the socio- technical system that – through complex system interdependencies – led to the emergence of new issues. The more momentum the socio-technical system gained, the more unforeseen technological developments exerted direct pressures on policy makers to adjust the design of previously implemented policies. In this sense, technological change served as both an outcome and a driver of policies targeted at inducing technological progress.
Based on the case study, the paper develops a descriptive framework of the cyclical interplay between policy design and technological change, based on Rosenberg’s (1969) ‘compulsive sequences’ in the development of technical systems. The framework, labeled ‘compulsive policymaking’ (shown in Figure 13 below), stresses the role of the dynamics and uncertainty of technological change and
29
describes how these factors affect deployment policy interventions. The implications of this framework for the design and governance of deployment policies is discussed in section 5.2 below.
Politics (b)
Prevalent Issues in Socio-Technical System (1)
System Limited Capacity Interdepen- and Foresight (a) Politics (b) dencies (e)
Policy-Makers‘ Technological Focus in Change (3) Policy Design (2)
Policy Intervention (c)
Other Influences on System (d)
Figure 13: The descriptive model of the cyclical interplay between policy design and technological change, labeled ‘compulsive policymaking’.
4.5. Paper 5: International Support for Feed-in Tariffs in Developing Countries –
A Review and Analysis of Proposed Mechanisms
Government support, in the form of feed-in tariff policies (FITs) for renewable electricity, has attracted large private-sector investments in sustainable electricity generation in the industrialized world (such as in Germany, as discussed in Paper 4). In an effort to replicate these experiences globally, a number of international organizations, NGOs, banks and donor countries are proposing mechanisms to cover part of the cost of FITs in developing countries. Paper 5 reviews these proposals for supported FITs and then uses a techno-economic model of Thailand’s Alternative Energy Development Plan 2013-2021 to analyze how feed-in tariffs (FITs) in developing countries can be supported through direct international financial assistance (Huenteler, 2014). A particular emphasis is placed on how policy design can ensure the long-term stability and effectiveness of such a policy.
Four main conclusions can be drawn from the analysis. First, the magnitude of the incremental cost of supported FITs is considerable. In the case of Thailand, the incremental cost of the FIT was estimated at USD 21bn or 3.15% of Thailand’s GDP in 2012. Second, the incremental costs of supported FITs in
30 Synopsis developing countries are very uncertain (e.g., depending on the fossil fuel price scenario the FIT may result in a cost of USD 22bn or savings of 31bn; see Figure 14). Third, the uncertainty in the incremental cost is driven, to a large extent, by the uncertain savings from avoided fossil fuel consumption. This is mostly due to the nature of the uncertainty. Unlike the payments to the supported renewable electricity generators, which are uncertain but can be managed by carefully designing the process in which commitments are made, the avoided cost uncertainty is resolved only over the lifetime of the supported projects. Fuel prices may vary substantially over the 10-20 years that FIT payments have to be committed, and changes over time in the type of displaced electricity may result in large step changes in the avoided cost. Fourth, the paper shows that the reviewed proposals for internationally supported FITs differ in how they allocate the avoided cost uncertainty between national governments and international donors – some proposals assign all cost uncertainty to the national government, whereas the international side assumes all uncertainty in others.
In billion USD2012, undiscounted IEA high-price scenario Medium-price scenario Low-price scenario 120
100 + 22.0 17.6 -21.2 80 -30.8
60
40
20
0 FIT M1 M2 M3 M4 M5 M6M7 M8 M9 M1 M10 M11 cost Average natural gas Average natural gas Average LNG Different price import price import price assumptions for displaced electricity
Figure 14: Avoided and incremental cost of a supported FIT in Thailand under different assumptions about fossil fuel cost scenarios and type of avoided electricity.
These findings have implications for the design of deployment policies in developing countries. With regard to supported FITs, the considerable uncertainty about policy cost implies challenges for the design and implementation of a supported FIT: donor countries will be unwilling to commit to financial assistance flows without knowing their eventual volume, while investors will only be attracted with a clear, long-term support commitment. This suggests that an international support mechanism that differentiates the allocation of uncertainty depending on the income level of the recipient country is more suitable for global-scale support than a one-size-fits-all approach: while emerging middle- income countries can be expected to absorb this risk, international donors might be willing assume it when they support the least-developed countries. More generally, Paper 5 offers valuable lessons for
31
governments in developing countries who aim to adopt specific deployment policy instruments that have proven successful in the developed world (see section 5.2).
4.6. Paper 6: Japan’s Post-Fukushima Challenge – Implications from the German
Experience on Renewable Energy Policy
Paper 6 describes some important lessons learned from the support of solar PV in Germany (see Paper 4) and discusses their implications in the context of Japanese solar policy after the Fukushima accident (Huenteler et al., 2012). As in paper 5, a particular emphasis is placed on how policy design can ensure that the policy remains stable and effective in the long term.
The paper first highlights the main challenges to Germany’s FIT. First, since about 2008, the mounting payment commitments of the scheme have fuelled a public and scientific debate about the scheme’s future (PV accounted for only 3% of electricity production while cumulative committed payments reached approximately €100bn in 2011). Second, evidence suggests that the generous FIT incentivized firms to reallocate resources to new production capacity and, in relative terms, away from R&D, which led to a declining R&D quota (see Figure 15 below). Third, as an industrial policy, Germany’s FIT was largely unsuccessful, as illustrated by the rising imports in Figure 15. Because of these factors, the legitimacy of support for PV eroded over time (eventually leading to the abolishment of the FIT in 2014).
Capacity additions and R&D intensity of PV industry net imports; blocks [in MW] in Germany; dotted line [in %]
8’000 7’408 3.5 3.0 6’000 2.5 2.0 4’000 3’806 1.5 1’809 1.0 2’000 1’271 670 951 843 0.5 42 78 118 139 0 0.0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Annual capacity additions [MW] Net imports (including shipments from German firms manufacturing abroad) [MW] R&D intensity: R&D investments per sales volume [%]
Figure 15: Main challenges for legitimacy of the FIT for PV in Germany: installations and net imports are growing exponentially while the research intensity in the industry is declining. For sources, see Paper 6.
Further, the paper discusses these challenges in the context of Japan’s new FIT policy for solar PV. It argues that concentration of regulatory authority for energy policy in the Ministry of Economy, Trade
32 Synopsis and Industry renders economic and industrial policy aspects of the FIT particularly important. Whenever plans for the Japanese energy sector and the diffusion of renewables have been drafted, they prominently featured industrial policy objectives. The new FIT, too, lists “promotion of the domestic industry, and thereby strengthening the international competitiveness of [Japan]” as one of its main targets. At the same time, changes in the global industry landscape might make it difficult for Japan to achieve such objectives. The growth of the global market allowed huge production capacities to be built up, increasingly located in low-wage countries. The cost and scale advantage of Chinese/Taiwanese firms poses a significant challenge for their Japanese counterparts.
The paper derives three implications. First, in the long run, the government should strive for an integrative policy framework, balancing priorities of energy security, environmental policy, climate policy, as well as economic and industrial policy. A process of ‘policy learning’ and refinement, as achieved in Germany over the last 10 years, is only possible under a political framework with balanced responsibilities. Second, institutionalizing a FIT revision process that involves several ministries, such as for the German FIT, could enhance transparency of the political process. Third, given its current political situation, Japan should aim to maintain policy legitimacy by fostering both rapid cost reductions and industrial competitiveness. The focus on market subsidies rather than research funding in Germany appears to have created incentives to favor manufacturing and scale effects over long-term research. One way forward could be to make FIT support conditional on efficiency criteria in order to incentivize both diffusion and investment in long-term R&D. A modified FIT could, for instance, require solar modules to fulfill a condition similar to one that had been implemented in an investment subsidy that was granted to residential systems in 2009: in order to receive the subsidy, conversion efficiency had to exceed the average on the market. Since Japan has a well-functioning innovation system in the semiconductor and solar PV industries, and is a high-wage country, it can be expected that such a policy is much better suited to the capabilities of the industry than a scheme merely rewarding production at the lowest costs.
5. Conclusions
Two sets of overarching conclusions can be drawn from the six papers of this thesis. The first set of conclusions relates to the question of how deployment policies in the energy sector can be designed to account for technological characteristics (RQ1; section 5.1). The second set of conclusions addresses
33
the way in which deployment policies can be designed to ensure effectiveness over the lifetime of the policy (RQ2; section 5.2).
5.1. Tailoring Deployment Policies to Specific Energy Technologies
Few scholars would support a 'one-size-fits-all' technology policy approach for semiconductor, machinery, biotechnology, oil and gas, and chemical industries. This thesis’ results indicate that it is equally misguided to lump together PV cells, wind turbines, biomass gasification, carbon capture and storage, and fuel cells when designing technology policy instruments. If the trillions of USD of public expenditures for deployment policies over the coming decades are to stimulate innovation effectively, their design needs to reflect the characteristics of the supported technologies.
5.1.1. Implications for Deployment Policy Design
The results presented in the first three papers demonstrate that technological characteristics – in particular, the product architecture, the cost structure and the scale of the production process – shape the temporal and spatial patterns of technological learning in energy technologies.
In terms of temporal patterns, the results suggest that the design of energy technologies that exhibit the characteristics of complex products and systems (complex product architectures and low production volume) continues to evolve over decades after the onset of large-scale deployment. In mass-produced energy technologies, the focus of most innovative activity is on the production process soon after the technology is deployed at scale. The two different patterns of learning imply different roles for deployment policies in the innovation process, and therefore have implications for deployment policy objectives and design, which are discussed below.
When going beyond the technologies analyzed in this thesis, it quickly becomes clear that the dichotomy of ‘complex products and systems’ and ‘mass produced technologies’ alone does not suffice to describe the full variety of energy technologies. Therefore, and given the continuous nature of the two determinants, the two life-cycle patterns are best understood as a continuum, as shown Figure 16. The higher the complexity of the product architecture and the smaller the scale of the production process, the more the policy implications of complex products and systems apply, and vice versa.
34 Synopsis
Experimentation and user-producer interaction Grants for innovative features High Consortia, private-public partnerships Domestic markets Performance competition Complexity of product Data collection and publication architecture Evolving standards Scale Large, ideally global markets Cost competition Reverse auctions Low Rapid adjustment of incentives
Low High Scale of production process
Figure 16: Objectives and design features of deployment policies for different types of energy technologies.
Deployment policies for technologies with a complex product architecture, such as wind turbines, geothermal systems, nuclear power plants, and tidal energy systems, deployment policies have to go beyond simply subsidizing scale to in order to fully realize their potential innovation impact. For these technologies, deployment policies need to be understood as R&D policies rather than merely as subsidies. Simply doing ‘more of the same’ will not stimulate innovation in these technologies. Rather than enabling economies of scale, deployment policies should be targeted at creating ‘performance- driven’ niche markets (Grubler and Wilson, 2014): they should not aim for very large roll-out of existing technologies, but be explicitly be targeted at reducing technological uncertainty. For example by providing grants for innovative technology features, technology platforms, public-private partnerships, or by financing experimentation in different geographical and climatic environments. Furthermore, deployment policies could be accompanied by measures to enhance user-producer interaction (e.g., technology platforms or grants for consortia), improve market transparency (through collecting and publishing performance data) and gradually adjust performance standards (e.g., as it has been done with grid-integration requirements for wind turbines).
Deployment policies for mass-produced technologies, such as solar PV, in contrast, need to be understood as means to create conditions for large-scale production. Large markets, ideally coordinated internationally, are needed to enable the necessary economies of scale and the learning-by-doing in production. At the same time, policy support needs to make sure that cost competition remains high, e.g., by auctioning off subsidies, or by requiring benefiting companies to reveal production cost data and adjusting incentivizes continuously.
35
From a spatial perspective, markets for mass-produced goods should ideally be supported globally, both because of the necessary market scale and because there are significant free-rider effects for countries that wait and deploy later. Markets for complex products and systems should be supported on a regional or national level to reap the benefits of learning, because a patchwork of small niche markets will not overcome the ‘chicken and egg’ problem of low production volumes and high production cost.
Figure 17 locates a broader set of energy technologies in the technology space generated by the two characteristics. Complex products and systems are further divided into infrastructure systems (such as public transport systems and electricity grids) and design-intensive products (which include most large electric power plants). Infrastructure systems refer to technologies that are highly complex and are provided through a project-based production process, and thus involve hardly any process innovation. Design-intensive products are produced in small but significant quantities and thus involve some form of process innovation. On the other end of the spectrum, mass-produced technologies are divided into continuous-flow processes (such as transport fuels), for which the process is the primary focus of innovation from beginning, and process-intensive products, which involve some experimentation with different product designs in the beginning. The graphic also shows two groups of technologies that do not fit on the continuum between mass-produced technologies and complex products and systems: (i) low-tech products (small wind, small hydro turbines) which are relatively simple and produced in very small batches, and have potential for neither significant product nor process innovation, and (ii) mass- produced complex products (electric cars, grid-scale batteries), which involve continued product and process innovations over the entire technology life-cycle.
Infrastructure systems Transport systems, High electricity grids
Design-intensive products Mass-produced complex products Gas turbines, wind Electric cars, grid-scale batteries Complexity of product architecture Low-tech products Process-intensive products Small hydro, small wind Solar PV, fuel cells
Low Continuous-flow processes Biofuels, building materials Low High
Scale of production process
Figure 17: Stylized classification of different energy technologies according to scale of production process and complexity of product architecture.
36 Synopsis
5.1.2. Implications for Countries’ Technology Strategies
Besides the design of deployment policies, the spatial patterns of technological learning also have important implications for the choice of technologies that countries support. This is particularly important for developing countries, which aim to attract local manufacturing but have only limited public resources to support the deployment of innovative energy technologies. Possible strategies for design-intensive and process-intensive types of technologies and three country types (low-, middle-, or high-income country) are listed in Table 3.
In complex products and systems, low- and middle-income countries have opportunities in the supply of components, such as mirrors for concentrating solar power plants (North Africa), parts for geothermal power plants (Indonesia) or towers for wind turbines (South Africa), which are often costly to transport. If the domestic market is large enough, prolonged experience with the supply of components for local projects may give firms a competitive edge that may lead to exports into other developing countries. Another field for domestic engagement for low- and middle-income countries is operation and maintenance, which can often become a significant share of value-add for design- intensive technologies. Middle-income countries may go beyond that and, with persistent domestic support over a long time, even become competitive system integrators in global markets, as both China and India are demonstrating in wind energy, and China in the field of large hydropower. To become competitive in the global market for final products, domestic firms need to engage in state-of-the-art technological activity over extended periods of time. That requires either early entry into the global market or very persistent domestic policy support. Likely only a few large countries outside the developed world (e.g., China or India) can afford such technology strategies.
Table 3: Technology strategies for different income-classes of countries.
Activity of domestic firms in…
Low-income country Middle-income country High-income country
Design-intensive energy Peripheral components, operation and Components, installation, operation System integration, core components technologies maintenance and maintenance
Simple production steps, installation Process-intensive energy Installation and/or Manufacturing equipment technologies Production and export
In mass-produced products, much of the required know-how for manufacturing can be acquired by purchasing production equipment from advanced economies (technology transfer in the semiconductor,
37
textile, and consumer durables industries took this path, for example). If they have access to large export markets, firms in middle-income countries can become globally competitive, since they often face lower unit costs in terms of labor and energy. Becoming a manufacturing hub for technologies such as solar PV, solar heating (vacuum tubes), heat pumps, energy-saving building materials, or energy-efficient lighting might thus be a suitable strategy for middle-income countries with access to large domestic or global markets (in the field of solar PV, Malaysia, the Philippines and especially China are recent successful examples). High-income countries often focus on the export of manufacturing equipment related to these technologies. Low-income countries, on the other hand, have neither components to focus on (since the products are rather simple and often small in size), nor the opportunity to engage significantly in operation and maintenance (which is usually rather simple and takes a small share of value-add).
5.2. Designing Deployment Policies that are Effective in the Long Run
The results presented in the second half of the thesis have three important implications for the design of deployment policy interventions.
First, the findings suggest that a critical task in designing effective policies targeted at fostering technological change lies not only in developing adequate policy instruments based on existing knowledge about the technology but also in investing resources into monitoring and understanding the technological dynamics of the system to be intervened in. The study stresses the potential that lies in developing a profound knowledge about processes related to technological change and leveraging this knowledge when designing corresponding policies. For example, according to learning curve theory, cost reduction in solar PV technology is a function of deployment, but German policy makers chose a degression mechanism that reduced the remuneration for newly installed plants as a function of time, which turned out to be inadequate when the rates of deployment grew exponentially. A more profound understanding of the dynamics underlying cost reductions at the time of policy development might have prevented many of the iterations we have seen. To be sure, due to the complex nature of socio- technical systems it is by no means possible to accurately foresee the outcome of policy interventions based solely on the analysis of historical cases. Therefore, we suggest making system analysis an integral part of policy monitoring. Rather than only tracking the outcome of policy interventions, policy makers should try to understand the root cause of unexpected technology dynamics and their relation to policy.
38 Synopsis
Second, the analysis also has implications for policy makers who wish to adopt policy instruments that are already successfully used in other countries. Our study shows that the effectiveness and appropriateness of policy instruments at any given time is at least partly conditional on previous policy interventions and resulting changes in the socio-technical system. This implies that, although policy makers should leverage the lessons learned in other countries, there is a limit to which policy features implemented in other countries can be successfully copied. For example, the diffusion of solar PV in the German socio-technical system may now have built up enough momentum, e.g., in the form of trust in the system and vested interests, that the system can wither policy interventions aimed at increasing the cost effectiveness or grid-friendliness of the policy support. Introducing measures such as the mandatory direct marketing of renewable power in a nascent system without the same history of PV deployment may not show the same positive effects as in the German context as they may induce investment uncertainty and derail the deployment of the technology before it has even picked up momentum.
39
6. Overview of the Papers
The six papers that form this thesis are included in the following chapters. They are included either as published by the journal or as submitted to the journal. Five of the six papers are co-authored. For each of these papers, Table 4 below lists the co-authors and their respective contribution. The submission status of the papers is as of November 30, 2014.
Table 4: Overview of the six papers included in this dissertation.
# Title Authors Authors’ contribution Status / Journal
How a Product’s Design Hierarchy Joern Huenteler, Jan JH and TS developed the model, JH and JO Shapes the Evolution of Technological Ossenbrink; Tobias prepared the data, JH conducted the analysis Under review at Research 1 Knowledge–Evidence from Wind Schmidt, Volker and interpreted the results, JH, JO, TS and Policy since 11/07/2014. Turbine Technology Hoffmann VH wrote the paper
Technology Life-Cycles in the Energy JH and TS developed the model, JH and JO Working paper to be Joern Huenteler, Tobias Sector – Technological Characteristics prepared the data, JH conducted the analysis submitted to Technological 2 Schmidt, Jan Ossenbrink; and the Role of Deployment for and interpreted the results, JH, JO, TS and Forecasting and Social Volker Hoffmann Innovation VH wrote the paper Change.
The Effect of Local and Global Learning Joern Huenteler, JH, CN, TS developed the model, JH In press, Journal of Cleaner 3 on the Cost of Renewable Energy in Christian Niebuhr, conducted the analysis and interpreted the Production. Developing Countries Tobias Schmidt results, JH and TS wrote the paper
JHH and JH developed the model, JHH Compulsive Policy-Making–The Joern H. Hoppmann, conducted the analysis, JHH, JH and BG Published in Research Policy 4 Evolution of the German Feed-in Tariff Joern Huenteler, Bastien interpreted the results, JHH, JH and BG 2014, 43 (8), p. 1422-1441. System for Solar Photovoltaic Power Girod wrote the paper
International Support for Feed-in Tariffs JH, CN, TS developed the model, JH Published in Renewable and 5 in Developing Countries–A Review and Joern Huenteler conducted the analysis and interpreted the Sustainable Energy Reviews Analysis of Proposed Mechanisms results, JH wrote the paper 2014 (11), 39, p. 857–873.
Japan’s Post-Fukushima Challenge– Joern Huenteler, Tobias JH and TS developed the idea, JH, TS and Published in Energy Policy 6 Implications From the German Schmidt, Norichika NK wrote the paper 2012, 45 (4), p. 6-11. Experience on Renewable Energy Policy Kanie
40 Synopsis
References
102nd Congress, 1992. Energy Policy Act of 1992 (Enrolled Bill [Final as Passed Both House and Senate]). Washington, DC.
AEIC, 2010. A Business Plan for America’s Energy Future. American Energy Innovation Council (AEIC), Washington, DC.
Anadon, L.D., Bunn, M., Chan, G., Chan, M., Jones, C., Kempener, R., Lee, A., Logar, N., Narayanamurti, V., 2011. Transforming US Energy Innovation. Belfer Center for Science and International Affairs, Cambridge, MA.
Andersen, P.D., 2004. Sources of experience? theoretical considerations and empirical observations from Danish wind energy technology. International journal of energy technology and policy 2, 33.
Anderson, P., Tushman, M.L., 1990. Technological Discontinuities and Dominant Designs: A Cyclical Model of Technological Change. Administrative Science Quarterly 35, 604–633.
Baldwin, C., Maccormack, A., Rusnak, J., 2014. Hidden Structure: Using Network Methods to Map Product Architecture. Research Policy 43, 1381–1397.
Beise-Zee, M., 2004. Lead markets: country-specific drivers of the global diffusion of innovations. Research Policy 33, 997–1018.
Bennett, C., Howlett, M., 1992. The lessons of learning: Reconciling theories of policy learning and policy change. Policy sciences 25, 275–294.
Bergek, A., Jacobsson, S., Sandén, B.A., 2008. “Legitimation” and “development of positive externalities”: two key processes in the formation phase of technological innovation systems. Technology Analysis & Strategic Management 20, 575–592.
Bettencourt, L.M. a, Trancik, J.E., Kaur, J., 2013. Determinants of the pace of global innovation in energy technologies. PloS one 8, e67864.
BNEF, 2014. Climatescope 2014. Bloomberg New Energy Finance (BNEF), Multilateral Investment Fund part of the Inter-American Development Bank, UK Department for International Development, Power Africa, New York, NY.
Campbell, D.T., 1990. Levels of organization, downward causation, and the selection-theory approach to evolutionary epistemology, in: Greenberg, G., Tobach, E. (Eds), Theories of the Evolution of Knowing. Lawrence Erlbaum, Hillsdale, NJ, pp. 1–17.
41
Clark, K.B., 1985. The interaction of design hierarchies and market concepts in technological evolution. Research Policy 14, 235–251.
Creswell, J., 2013. Research design: Qualitative, quantitative, and mixed methods approaches. Sage Publications, Thousand Oaks, CA; London, UK; New Delhi, India.
Dedehayir, O., Mäkinen, S., 2011. Determining reverse salient types and evolutionary dynamics of technology systems with performance disparities. Technology Analysis & Strategic Management 23, 37–41.
Dosi, G., 1982. Technological paradigms and technological trajectories: A suggested interpretation of the determinants and directions of technical change. Research Policy 11, 147–162.
Edenhofer, O., Madruga, R.P., Sokona, Y., Seyboth, K., Matschoss, P., Kadner, S., Zwickel, T., Eickemeier, P., Hansen, G., Schlömer, S., von Stechow, C., 2012. Renewable Energy Sources and Climate Change Mitigation. Special Report of the Intergovernmental Panel on Climate Change (IPCC).
Edler, J., 2013. Review of Policy Measures to Stimulate Private Demand for Innovation: Concepts and Effects, in: Compendium of Evidence on the Effectiveness of Innovation Policy Intervention, Available Online at Http://innovation-Policy.net/compendium/. Manchester Institute of Innovation Research (MIoIR), University of Manchester, Manchester, UK.
Edler, J., Georghiou, L., 2007. Public procurement and innovation-Resurrecting the demand side. Research Policy 36, 949–963.
E-FI, 2014. Report on Research, Innovation and Technological Performance in Germany. Expert Commission on Research and Innovation (E-FI), Berlin, Germany.
Ethiraj, S., 2007. Allocation of inventive effort in complex product systems. Strategic Management Journal 584, 563–584.
Fagerberg, J., Fosaas, M., Sapprasert, K., 2012. Innovation: Exploring the knowledge base. Research Policy 41, 1132–1153.
Fagerberg, J., Verspagen, B., 2009. Innovation studies - The emerging structure of a new scientific field. Research Policy 38, 218–233.
Fleck, J., 1988. Innofusion or diffusation?: The nature of technological development in robotics. Edinburgh PICT Working Paper No. 4.
Foray, D., Mowery, D.C., Nelson, R.R., 2012. Public R&D and social challenges: What lessons from mission R&D programs? Research Policy 41, 1697–1702.
42 Synopsis
Friedlingstein, P., Andrew, R.M., Rogelj, J., Peters, G.P., Canadell, J.G., Knutti, R., Luderer, G., Raupach, M.R., Schaeffer, M., van Vuuren, D.P., Le Quéré, C., 2014. Persistent growth of CO2 emissions and implications for reaching climate targets. Nature Geoscience 7, 709–715.
Frondel, M., Ritter, N., Schmidt, C.M., 2008. Germany’s solar cell promotion: Dark clouds on the horizon. Energy Policy 36, 4198–4204.
Frondel, M., Ritter, N., Schmidt, C.M., Vance, C., 2009. Economic impacts from the promotion of renewable energy technologies: The German experience. Energy Policy 38, 4048–4056.
Galiana, I., Green, C., 2009. Let the global technology race begin. Nature 462, 570–571.
Gallagher, K., Grubler, A., Kuhl, L., Nemet, G., Wilson, C., 2012. The Energy Technology Innovation System. Annual Review of Environment and Resources 37, 137–162.
Gallagher, K.S. (Ed.), 2009. Acting in Time on Energy Policy. Brookings Institution, Washington, DC.
Gallagher, K.S., 2014. The Globalization of Clean Energy Technology. MIT Press, Cambridge, MA.
Gallagher, K.S., Anadon, L.D., Kempener, R., Wilson, C., 2011. Trends in investments in global energy research, development, and demonstration. Wiley Interdisciplinary Reviews: Climate Change 2, 373–396.
GAO, 2013. Report to Congressional Requesters - Wind Energy: Additional Actions Could Help Ensure Effective Use of Federal Financial Support. United States Government Accountability Office, Washington, DC.
Garrad, A., 2012. The lessons learned from the development of the wind energy industry that might be applied to marine industry renewables. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences 370, 451–71.
German Federal Diet, 1999. Plenary Protocol 14/39 (in German). accessed online at dip21.bundestag.de/dip21/btp/14/14039.pdf.
Gillingham, K., Sweeney, J., 2010. Market Failure and the Structure of Externalities. Harnessing Renewable Energy in Electric Power Systems: Theory, Practice, Policy 69.
Gipe, P., 1995. Wind energy comes of age. John Wiley & Sons Inc, New York, NY.
Hayami, Y., Ruttan, V., 1973. Professor Rosenberg and the Direction of Technological Change: A Comment. Economic Development and Cultural Change 21, 352–355.
Henderson, R.M., Clark, K.B., 1990. Architectural Innovation: The Reconfiguration of Existing Product Technologies and the Failure of Established Firms. Administrative Science Quaterly 35, 9–30.
43
Henderson, R.M., Newell, R.G., 2011. Accelerating Energy Innovation - Insights from Multiple Sectors. The Chicago University Press, Chicago, IL.
Hoppmann, J., Huenteler, J., Girod, B., 2014. Compulsive policy-making—The evolution of the German feed- in tariff system for solar photovoltaic power. Research Policy 43, 1422–1441.
Hoppmann, J., Peters, M., Schneider, M., Hoffmann, V.H., 2013. The two faces of market support—How deployment policies affect technological exploration and exploitation in the solar photovoltaic industry. Research Policy 42, 989–1003.
Huenteler, J., 2014. International support for feed-in tariffs in developing countries—A review and analysis of proposed mechanisms. Renewable and Sustainable Energy Reviews.
Huenteler, J., Niebuhr, C., Schmidt, T.S., 2014a. The effect of local and global learning on the cost of renewable energy in developing countries. Journal of Cleaner Production (in press).
Huenteler, J., Ossenbrink, J., Schmidt, T.S., Hoffmann, V.H., 2014b. How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Patent-Citation Networks in Wind Power. Working Paper available online at http://scholar.harvard.edu/files/jhuenteler/files/huenteler_et_al_design_hierarchy.pdf.
Huenteler, J., Schmidt, T.S., Kanie, N., 2012. Japan’s post-Fukushima challenge – implications from the German experience on renewable energy policy. Energy Policy 45, 6–11.
Huenteler, J., Schmidt, T.S., Ossenbrink, J., Hoffmann, V.H., 2014c. Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation. Working Paper.
Hughes, T.P., 1992. The dynamics of technological change: salients, critical problems, and industrial revolutions, in: Dosi, G., Giannetti, R., Toninelli, P.M. (Eds), Technology and Enterprise in A Historical Perspective. Clarendon Press, Oxford, UK.
IEA, 2013. Tracking Clean Energy Progress 2013.
IEA, 2014a. World Energy Outlook 2014 World Energy Outlook. International Energy Agency (IEA), Paris, France.
IEA, 2014b. CO2 Emissions from Fuel Combustion. International Energy Agency (IEA), Paris, France.
IEA, 2014c. IEA Energy Technology R&D Statistics - RD&D Budget IEA Energy Technology R&D Statistics (online database). Paris, France.
44 Synopsis
IPCC, 2007. Climate Change 2007: Mitigation of Climate Change. Contribution of Working Group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.
IPCC, 2013. Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.
IPCC, 2014. Climate Change 2014 Synthesis Report. Intergovernmental Panel for Climate Change (IPCC) Fifth Assessment Synthesis Report, Geneva, Switzerland.
Jacobsson, S., Bergek, A., 2004. Transforming the energy sector: the evolution of technological systems in renewable energy technology. Industrial and Corporate Change 13, 815–849.
Jacobsson, S., Lauber, V., 2005. Germany: from a modest feed-in law to a framework for transition, in: Lauber, V. (Ed), Switching to Renewable Power: A Framework for the 21 Century. Routledge, London, UK, pp. 122–158.
Jacobsson, S., Lauber, V., 2006. The politics and policy of energy system transformation---explaining the German diffusion of renewable energy technology. Energy Policy 34, 256–276.
Jacobsson, S., Sandén, B.A., Bångens, L., 2004. Transforming the Energy System—the Evolution of the German Technological System for Solar Cells. Technology Analysis & Strategic Management 16, 3–30.
Jaffe, A.B., Newell, R.G., Stavins, R.N., 2005. A tale of two market failures: Technology and environmental policy. Ecological Economics 54, 164–174.
JAIF, 2012. Nihon no genshiryokuhatsudensho no untenjoukyou (Service status of Nuclear Power Plant in Japan). Japan Atomic Industrial Forum, Inc (JAIF); published online on 2012/02/03, available at http://www.jaif.or.jp/ja/news/news_detail.php.
Johnstone, N., Haščič, I., Popp, D., 2010. Renewable energy policies and technological innovation: Evidence based on patent counts. Environmental and Resource Economics 45, 133–155.
Kageyama, Y., 2012. Japan plans to scrap nuclear plants after 40 years to beef up safety after Fukushima disaster. Associated Press on 2012/01/07.
Karnøe, P., 1993. Approaches to innovation in modern wind energy technology: technology policies, science, engineers and craft traditions. Center for Economic Policy Research Publication 334.
Kempener, R., Anadon, L.D., Gallagher, K.S., Jiang, K., 2012. Energy RD&D Investments in the Major Emerging Economies and the United States. Belfer Center for Science and International Affairs, Cambridge, MA.
45
Kline, S.J., Rosenberg, N., 1986. An overview of innovation, in: Landau, R., Rosenberg, N. (Eds), The Positive Sum Strategy: Harnessing Technology for Economic Growth. National Academic Press, Washington, DC, pp. 275–305.
Kossoy, A., Guigon, P., 2012. State and Trends of the Carbon Market 2012. The World Bank, Washington, DC.
Laleman, R., Albrecht, J., 2014. Comparing push and pull measures for PV and wind in Europe. Renewable Energy 61, 33–37.
Lindblom, C.E., 1959. The science of “muddling through.” Public administration review 19, 79–88.
Loftus, P., Cohen, A., 2014. A critical review of global decarbonization scenarios: what do they tell us about feasibility? Wiley Interdisciplinary Reviews: Climate Change (in press; doi:10.1002/wcc.324).
Malerba, F., 2002. Sectoral systems of innovation and production. 31, 247–264.
Markard, J., 2011. Transformation of infrastructures: sector characteristics and implications for fundamental change. Journal of Infrastructure Systems 17, 107.
Mathews, J. a, Tan, H., 2014. Economics: Manufacture renewables to build energy security. Nature 513, 166–8.
May, P., 1992. Policy learning and failure. Journal of public policy 12, 331–354.
Mazzucato, M., 2011. The Entrepreneurial State: Debunking Public vs. Private Sector Myths Soundings. Anthem Press, London, UK.
McNerney, J., Farmer, J., Redner, S., Trancik, J.E., 2011. Role of design complexity in technology improvement. Proceedings of the National Academy of Sciences 108, 9008–9013.
Mowery, D.C., 2012. Defense-related R&D as a model for “Grand Challenges” technology policies. Research Policy 41, 1703–1715.
Mowery, D.C., Nelson, R.R., Martin, B.R., 2010. Technology policy and global warming: Why new policy models are needed (or why putting new wine in old bottles won’t work). Research Policy 39, 1011–1023.
Mowery, D.C., Rosenberg, N., 1979. The influence of market demand upon innovation: a critical review of some recent empirical studies. Research Policy 8, 102–153.
Murmann, J.P., Frenken, K., 2006. Toward a systematic framework for research on dominant designs, technological innovations, and industrial change. Research Policy 35, 925–952.
Negro, S.O., Alkemade, F., Hekkert, M.P., 2012. Why does renewable energy diffuse so slowly? A review of innovation system problems. Renewable and Sustainable Energy Reviews 16, 3836–3846.
46 Synopsis
Nelson, R.R., Winter, S.G., 1977. In search of useful theory of innovation. Research Policy 6, 36–76.
Nemet, G.F., 2009. Demand-pull, technology-push, and government-led incentives for non-incremental technical change. Research Policy 38, 700–709.
Nemet, G.F., Baker, E., Barron, B., Harms, S., 2014. Characterizing the effects of policy instruments on the future costs of carbon capture for coal power plants. Working Paper available at https://www.diw.de/documents/dokumentenarchiv/17/diw_01.c.467791.de/nemet_ams_jul2014_full- paper.pdf.
Nemet, G.F., Kammen, D.M., 2007. US energy research and development: Declining investment, increasing need, and the feasibility of expansion. Energy Policy 35, 746–755.
Norberg-Bohm, V., 2000. Creating Incentives for Environmentally Enhancing Technological Change: Lessons From 30 Years of US Energy Technology Policy. Technological Forecasting and Social Change 65, 125– 148.
PCAST, 1997. Report to the President on Federal Energy Research and Development for the 21st Century. The President's Committee of Advisors on Science and Technology (PCAST), Washington, DC.
PCAST, 1999Report to the President on Powerful Partnerships: The Federal Role in International Cooperation on Energy Innovation. The President's Committee of Advisors on Science and Technology (PCAST), Washington, DC.
Peters, M., Schneider, M., Griesshaber, T., Hoffmann, V., 2012. The impact of technology-push and demand- pull policies on technical change–Does the locus of policies matter? Research Policy 41, 1296–1308.
Ragwitz, M., Walz, R., Hoffmann, V., Schmidt, T., Neuhoff, K., Cantner, U., Graf, H., Wüstenhagen, R., Truffer, B., Schleich, J., Haas, R., Hekkert, M., Negro, S., Jacobsson, S., Kern, F., Rogge, K., 2014. Wirkung des EEG – was ist die empirische Evidenz? (Impacts of the Renewable Energy Law – What is the Empirical Evidence?). available online at http://www.isi.fraunhofer.de/isi- de/service/presseinfos/2014/pri-04-2014-EFI-Expertenstatement.php.
Rosenberg, N., 1969. The direction of technological change: inducement mechanisms and focusing devices. Economic Development and Cultural Change 18, 1–24.
Rosenkopf, L., Nerkar, A., 1999. On the Complexity of Technological Evolution - Exploring Coevolution Within and Across Hierarchical Levels in Optical Disc Technology, in: Baum, J.A.C., McKelvey, B. (Eds), Variations in Organization Science: In Honor of D. T. Campbell. SAGE Publications, Inc, New York, NY.
Sagar, A.D., van der Zwaan, B.C.C., 2006. Technological innovation in the energy sector: R&D, deployment, and learning-by-doing. Energy Policy 34, 2601–2608.
47
Sanderson, I., 2002. Evaluation, policy learning and evidence‐based policy making. Public administration 80, 1– 22.
Sanford, T., Frumhoff, P.C., Luers, A., Gulledge, J., 2014. The climate policy narrative for a dangerously warming world. Nature Climate Change 4, 164–166.
Saviotti, P., 1986. Systems theory and technological change. Futures 6, 773–786.
Schmidt, T., Schneider, M., 2012. The effects of climate policy on the rate and direction of innovation: A survey of the EU ETS and the electricity sector. Environmental Innovation and Societal Transitions Innovation 2, 23–48.
Shellenberger, M., Nordhaus, T., Hayward, S., 2010. Post-Partisan Power. The Breakthrough Institute, Washington, DC.
Simon, H., 1962. The architecture of complexity. Proceedings of the American Philosophical Society 106, 467– 482.
Taylor, M., 2012. Innovation under cap-and-trade programs. Proceedings of the National Academy of Sciences 109, 4804–4809.
Taylor, M.R., Rubin, E.S., Hounshell, D.A., 2005. Control of SO2 emissions from power plants: A case of induced technological innovation in the U.S. Technological Forecasting and Social Change 72, 697–718.
The White House, 2009. The Recovery Act. Washington, DC.
The White House, 2014. U.S.-China Joint Announcement on Climate Change. The White House, Washington, DC, available online at http://www.whitehouse.gov/the-press-office/2014/11/11/us-china- joint-announcement-climate-change.
The World Bank, 2013. Annual Report 2013. The World Bank, Washington, DC.
Tilman, D., Hill, J., Lehman, C., 2006. Carbon-negative biofuels from low-input high-diversity grassland biomass. Science (New York, N.Y.) 314, 1598–600.
Trancik, J.E., 2014. Back the renewables boom. Nature 507, 4–6.
Truffer, B., Markard, J., Binz, C., Jacobsson, S., 2012. Energy Innovation Systems: Structure of an emerging scholarly field and its future research directions. Working Paper available at http://www.eis- all.dk/~/media/Sites/EIS_Energy_Innovation_Systems/english/about%20eis/Publications/EIS_Radarpape r_final.ashx.
Tushman, M., Rosenkopf, L., 1992. Organizational determinants of technological-change-toward a sociology of technological evolution. Research in Organizational Behavior 14, 311.
48 Synopsis
UNEP, 2014. The emissions gap report 2014. United Nations Environmental Program (UNEP) Finance Initiative, Nairobi, Kenya.
UNFCCC, 1992. United Nations Framework Convention. United Nations Framework Convention on Climate Change (UNFCCC), Bonn, Germany.
UNFCCC, 2009. Copenhagen Accord. United Nations Framework Convention on Climate Change (UNFCCC) - Conference of the Parties, Copenhagen, Denmark.
UNFCCC, 2014. Quantified economy-wide emission reduction targets by developed country Parties to the Convention: assumptions, conditions, commonalities and differences in approaches and comparison of the level of emission reduction efforts. United Nations Framework Convention on Climate Change (UNFCCC), Bonn, Germany.
Unruh, G.C., 2000. Understanding carbon lock-in. Energy Policy 28, 817–830.
Walker, W., 2000. Entrapment in large technology systems: institutional commitment and power relations. Research Policy 29, 833–846.
Wene, C., 2000. Experience curves for energy technology policy. International Energy Agency (IEA) / Organisation for Economic Cooperation and Development (OECD), Paris, France.
Weyant, J.P., 2011. Accelerating the development and diffusion of new energy technologies: Beyond the “valley of death.” Energy Economics 33, 674–682.
Wiesenthal, T., Leduc, G., Haegeman, K., Schwarz, H.-G., 2011. Bottom-up estimation of industrial and public R&D investment by technology in support of policy-making: The case of selected low-carbon energy technologies. Research Policy 41, 116–131.
Wilson, C., 2012. Up-scaling, formative phases, and learning in the historical diffusion of energy technologies. Energy Policy 50, 81–94.
Winskel, M., Markusson, N., Jeffrey, H., 2014. Learning pathways for energy supply technologies: Bridging between innovation studies and learning rates. Technological Forecasting & Social Change 81, 96–114.
Yang, C.-J., Oppenheimer, M., 2007. A “Manhattan Project” for climate change? Climatic Change 80, 199–204.
Zhang, X., Karplus, V.J., Qi, T., Zhang, D., He, J., 2014. Carbon emissions in China: How far can new efforts bend the curve? MIT Joint Program on the Science and Policy of Global Change, Report No. 267 October 2014, Cambridge, MA.
49
How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Patent-Citation Networks in Wind Power †
Joern Huenteler1,2*, Jan Ossenbrink1, Tobias S. Schmidt1,3, Volker H. Hoffmann1
1Department of Management, Technology and Economics, ETH Zurich, Switzerland 2Belfer Center for Science and International Affairs, John F. Kennedy School of Government, Harvard University, USA 3Precourt Energy Efficiency Center, Stanford University, USA *corresponding author: [email protected]; +41 44 632 97 39; +41 44 632 05 41
†Submitted to Research Policy on October 23, 2014 (under review since November 7, 2014). Previous versions of this paper have been presented at the Science, Technology and Public Policy seminar at Harvard in 2013, the International Sustainability Transitions 2013 conference at ETH Zurich and the Energy Systems in Transition 2013 conference at Karlsruhe Institute of Technology.
Abstract
We analyze how a product's design hierarchy shapes the evolution of the underlying body of technological knowledge, building on the literature on technological evolution in complex products. This literature suggests that the design hierarchy of a product can have an ordering effect on the evolution of commercialized artifacts, in particular when product design decisions on high levels of the design hierarchy set the agenda for subsequent variation and experimentation on lower levels. We extend this literature by analyzing the design hierarchy's effect on the evolution of the industry's knowledge base, using the case of wind turbine technology over the period 1973-2009. We assess the technological focus of patents along the core trajectory of knowledge generation, identified through a patent-citation network analysis, and link it to a classification of technological problems into different levels in the design hierarchy. Our analysis suggests that the evolution of an industry's knowledge base along a technological trajectory is not a unidirectional process of gradual refinement: the focus of knowledge generation shifts over time between different sub- systems in a highly sequential pattern, whose order is strongly influenced by the design hierarchy. Each of these shifts initiates a new process of integration of industry-external knowledge into the knowledge base, thus opening windows of competitive opportunity for potential entrants with strong knowledge positions in the sub-system that has moved into the focus of innovation. We discuss implications for the debate on supply-side and demand-side influences along technological trajectories and for the understanding of the competitive advantage of specific knowledge positions of firms and nations.
Keywords: Knowledge dynamics, Technological trajectory, Design hierarchy, Product architecture, Citation-network analysis, Wind power
52 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Abstract
We develop a methodology to study knowledge evolution along trajectory We analyze how focus of patenting in wind turbine technology shifted over time We demonstrate that sequence of focus in patenting follows core-periphery pattern Each shift in focus initiates new wave of integration of industry-external knowledge Our findings can explain shifts in competitive landscape along trajectory
53
1. Introduction
Complex, systemic products, such as power plants, aircraft and telecommunication networks, are a key entry channel for new technology into the economy (Rosenberg, 1963). Some consider them the ‘frontier’ of the economic development of nations (Hidalgo et al., 2007). They also underpin those sectors – manufacturing, energy, trade and transport – that are at the heart of the world’s environmental challenges. Technological change in such products takes the form of incremental innovation along established technological trajectories (Constant, 1973; Dosi, 1982; Clark, 1985; Frenken, 2006). Understanding the factors that shape the trajectories of technological evolution in complex products is therefore critical for technology strategies as well as economic and environmental policy (Acha et al., 2004; Davies and Hobday, 2005).
A number of qualitative studies emphasize the influence of the hierarchy of design decisions, or design hierarchy, on technological trajectories in complex products (e.g., Hughes, 1983; Clark, 1985; Vincenti, 1990). In particular, evidence suggests that movement along the technological trajectory in complex products is associated with movement down the design hierarchy in two principal ways: First, after a new trajectory has emerged, decisions about the overall product design often ‘set the agenda’ for subsequent change in sub-systems and individual components (Clark, 1985; Murmann and Tushman, 2002; Murmann and Frenken, 2006). Second, changes in sub-systems that perform the core functions of the product tend to precede changes in more peripheral sub-systems (Abernathy and Clark, 1985; Murmann and Frenken, 2006; Lee and Berente, 2013). The movement along the trajectory and down the design hierarchy implies change in the universe of commercialized designs – i.e., the evolution of artifacts – and in the underlying technological understanding – i.e., the evolution of knowledge (Dosi, 1982; Martinelli, 2012). The two are linked but are far from congruent: significant changes in artifacts may be the result of incremental gains of knowledge, and seemingly small changes in artifacts may require large changes in the underlying knowledge base (Funk, 2009; Martinelli, 2012). However, quantitative work on the structuring effect of the design hierarchy on technological trajectories has focused primarily on innovation and the evolution of artifacts (e.g., Saviotti and Trickett, 1992; Frenken et al., 1999; Frenken, 2006; Castaldi et al., 2009; Mendonça, 2012). With few exceptions (Rosenkopf and Nerkar, 1999; Lee and Berente, 2013), the influence of the design hierarchy on invention and the evolution of knowledge have received little attention.
To address this gap, we analyze how a product’s design hierarchy influences the evolution of knowledge. We do so in order to investigate the assumption that the development of an industry’s
54 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology knowledge base along the trajectory is predominantly a process of incremental growth and refinement, without cyclical or sequential changes in the focus of inventive activity and the importance of industry- external knowledge. On the one hand, it is commonly assumed that movement down the hierarchy leads to the entrenchment of existing knowledge positions, thus enhancing the competitive advantage of incumbent firms and nations through incremental knowledge growth and refinement, whereas movement up the hierarchy – through the creation of new trajectories – is associated with novel skills and expertise, thus opening windows of opportunity for new entrants (Abernathy and Clark, 1985; Henderson and Clark, 1990; Bekkers and Martinelli, 2012). On the other hand, how the focus of innovation shifts along the technological trajectory has also been at the heart of the more recent debate on the value of supply and demand-side subsidies for stimulating innovation in emerging clean technologies. Because demand-side subsidies are assumed to stimulate movement along existing technological trajectories, recent studies have argued that incentives to deploy technologies such as wind and solar power can be expected to lead to the exploitation and refinement of the existing knowledge base rather than to the exploration of new and potentially more radical solutions (Menanteau, 2000; Nemet, 2009; Hoppmann et al., 2013). A better understanding of how an industry’s knowledge base evolves along the trajectory can thus contribute to improved managerial and policy decisions.
In analyzing how a product’s design hierarchy influences the evolution of knowledge, this paper links two streams of literature: research on dominant designs and technological evolution in systemic artifacts (e.g., Frenken and Nuvolari, 2004; Murmann and Frenken, 2006; Mendonça, 2012) and research on trajectories of knowledge generation (e.g., Fontana et al., 2009; Barberá-Tomás et al., 2011; Epicoco, 2013). In particular, we develop a novel methodology that combines the manual, categorical analysis of commercialized designs, as employed in studies of dominant designs and technological evolution in systemic artifacts, with patent-citation network analysis, as employed in the literature on knowledge trajectories. This methodology allows us to bridge the artifact and knowledge dimensions by studying the influence of the design hierarchy, which derives from relationships between elements of the physical artifact, on the trajectory of knowledge generation in the industry. We apply this novel methodology to the case of wind turbine technology in the period 1973-2009.
The paper makes several distinct contributions to theory and methodology. Theoretically, we contribute to the literature on knowledge positions and competitive advantage (Bekkers and Martinelli, 2012; Epicoco, 2013; Choi and Anadon, 2014) and the literature on the impact of demand-side subsidies on R&D (Menanteau, 2000; Nemet, 2009; Hoppmann et al., 2013). Our findings suggest
55
that the evolution of an industry’s knowledge base along the technological trajectory is not a unidirectional process of gradual refinement but a sequential process that is structured by the design hierarchy: the focus of knowledge generation shifts over time between different sub-systems, with each shift initiating a new cycle of integration of industry-external knowledge into the knowledge base, a pattern we call creative sequences. Methodologically, our analysis contributes to recent efforts to identify linkages and linking mechanisms between the evolution of knowledge and the evolution of artifacts (Ethiraj, 2007; Barberá-Tomás et al., 2011; Bakker et al., 2012; Martinelli, 2012). We extend the methodology developed by Verspagen (2007) and others to study the knowledge and the artifact dimensions of technological trajectories in an integrated way, which may facilitate a deeper understanding of the interaction between the two domains.
In the following, Section 2 lays out the paper’s theoretical perspective and reviews the literature on technological evolution in systemic artifacts. Section 3 introduces the case of wind turbine technology and Section 4 presents the data sources and methodology. The results are presented in Section 5 and discussed in Section 6. Conclusions are summarized in Section 7.
2. Theoretical Perspective
Complex products are conceptualized in this paper as systemic artifacts (Saviotti, 1986; Tushman and Rosenkopf, 1992), consisting of a non-trivial number of interdependent sub-systems and components that jointly enable the system to perform a number of functions, or service characteristics. The sub- systems and components are organized by a product architecture, which allocates system functions to the individual components and defines the interfaces between them (Simon, 1962; Clark, 1985; Baldwin et al., 2014).
Technological evolution in complex products is understood as proceeding predominantly along technological trajectories through refinement within and extension of existing product architectures (Constant, 1973; Dosi, 1982; Frenken, 2006). When referring to sequences in the focus of innovation in the following subsections, we are concerned with incremental innovations along such trajectories.
2.1. The Sequential Pattern of Innovation in Systemic Artifacts
Historians of technology have noted the existence of sequential patterns of innovation in the evolution of technological artifacts (Rosenberg, 1969; Constant, 1980; Hughes, 1983; Vincenti, 1990). In this
56 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology context, sequential means that technological progress is concentrated in only a small fraction of a product’s components and possible directions of change, and that the focus of this concentration shifts over time between technological problems. The observed sequential pattern also implies that the focus of innovative activity is at least partly collective, in the sense that it can be observed on the level of communities of practitioners rather than individual problem-solvers or firms.
Langes (1969) observed that since the industrial revolution, innovations in technological systems have followed a challenge-response pattern in which technological breakthroughs call forth further, complementary innovations. He described for instance how Kay’s flying shuttle (1733), which allowed the development of automatic looms, was followed by rapid development of new spinning devices from the 1750s to the 1770s that supplied yarn more rapidly (Langes, 1969, p. 84).7 More generally, several studies have observed that the focus of innovative activity is often on those elements that keep other parts of the system from exploiting their full performance potential, and that new bottlenecks can arise in related components once such performance bottlenecks are resolved (Hughes, 1983, 1992; Sahal, 1985; Ethiraj, 2007; Dedehayir and Mäkinen, 2011). Rosenberg (1969, p. 111) used the term compulsive sequences to describe this self-generating, cyclical nature of problem-solving in systemic artifacts.
2.2. The Influence of the Design Hierarchy on the Evolution of Artifacts
While many had observed the sequential nature of technological change, Clark (1985) first described in detail what determines the sequence of innovative activity among the elements of a systemic artifact. The sequence of innovations in the automotive and semiconductor sectors in their early decades, he argued, can be understood as the outcome of two factors: the hierarchical organization of design decisions on the supply side and the gradual refinement of consumer preferences on the demand side. Murmann and Frenken (2006) integrated these two factors into one model that uses the term design hierarchy to capture the supply and demand side influences on the evolution of systemic artifacts.
7 To explain the challenge-response pattern, some economists have invoked induced changes in the relative prices of component technologies or input factors, e.g. the price of yarn in Kay’s flying shuttle (Hayami and Ruttan, 1973). Yet, many others have pointed out that as long as the cost of R&D is uncertain, a change in relative factor prices by itself cannot explain the highly selective focus of innovative activity in technological systems (Rosenberg, 1969; Mowery and Rosenberg, 1979; Dosi, 1982).
57
The design hierarchy locates each element in the system in two hierarchies, which jointly affect the evolution of systemic artifacts (see Figure 1): the hierarchy of nested parts, which locates the element in the hierarchy of systems, sub-systems, components, sub-components, and so on defined by the product architecture; and the hierarchy of control, which orders the elements on each level of the hierarchy of nested parts according to their relative importance for the demanded service characteristics – i.e., the principal categories of variables that underpin consumer choices, such as the speed, cost, noise and visual appearance of a car.
Hierarchy of nested parts
System
Sub-system 1
Hierarchy of Hierarchy of Component 1 control control Component 2 Component 3 Sub-system 2
Hierarchy of Component 1 control Component 2 Component 3
Figure 1: Two dimensions of the design hierarchy of systemic artifacts: the hierarchy of nested parts and the hierarchy of control.
How the hierarchy of control and the hierarchy of nested parts relate to the product architecture and service characteristics is shown in detail in Figure 2. The hierarchy of nested parts reflects the product architecture (arrow a in Figure 2) (Murmann and Tushman, 2002). It captures the tendency of the focus of innovative activity to shift over time from the system-level to sub-systems and components – i.e., from the general to the specific – as certain high-level design decisions set the agenda for incremental problem-solving efforts on lower levels. For instance, design decisions in the combustion chamber component of a piston-driven internal combustion engine have to build on (and thus succeed) system-level design decisions on the type of energy conversion (internal or external combustion) and energy transmission (piston or rotary internal combustion engines).8
8 In evolutionary theory, this effect is referred to as downward causation (Campbell, 1990; Rosenkopf and Nerkar, 1999).
58 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
(f)
Design hierarchy Sequences in the Product (a) evolution of architecture Hierarchy of nested (c) artifacts parts
(e1) (e2) (b1) Service Hierarchy of (d) Sequences in the characteristics (b ) control evolution of 2 knowledge
Figure 2: The influence of the design hierarchy on sequences in the evolution of artifacts and the evolution of knowledge (focus of this analysis is marked in grey).
The hierarchy of control reflects the interplay between the product architecture and the service characteristics (arrows b1 and b2 in Figure 2). It captures the effect that even within sub-systems and within components, some design decisions are more important than others and therefore have a controlling influence on them. In particular, when a new trajectory emerges, innovative activity first tends to focus on ‘core’ sub-systems and components that are most relevant to the service characteristics of a product. Later it shifts toward more ‘peripheral’ elements that facilitate the adaptation of certain service characteristics to newly emerging market segments (Lancaster, 1979; Teubal, 1979; Clark, 1985; Saviotti, 1996; Frenken et al., 1999). The focus of innovative activity in the early years of the automobile industry, for example, moved over time from the engine and the steering device to the transmission system, the chassis and other parts of the system (Clark, 1985).
The Murmann-Frenken model predicts that the evolution of artifacts is determined by the joint influence of the hierarchy of control and the hierarchy of nested parts (arrow c in Figure 2).9
2.3. The Influence of the Design Hierarchy on the Evolution of Knowledge
Innovation is a process that links the knowledge and artifact dimensions of technological trajectories (arrows e1 and e2 in Figure 2). However, the literature on the influence of the design hierarchy on technological evolution has treated the underlying body of knowledge mostly as a black box. Below we
9 Arrow f in Figure 2 is outside the scope of this paper; it captures the effect that in the long-run, incremental innovations along the trajectory can endogenously give rise to new trajectories if innovations and their diffusion in the market alter the demanded service characteristics (Levinthal, 1998) or create opportunities to change the prevailing product architecture (Henderson and Clark, 1990; Funk, 2009).
59
analyze how the design hierarchy affects the evolution of the knowledge base of an industry – and thus the value of different knowledge positions relative to the core of the trajectory. In particular, we aim to explore whether the Murmann-Frenken model is useful also in conceptualizing how the focus of knowledge generation changes over time as an industry moves along a technological trajectory. In this process, does the trajectory of knowledge move from the general to the specific and from the core to the periphery?
Recent studies provide fragmented evidence that the trajectory of knowledge evolution does reflect the design hierarchy. On a general level, Martinelli (2012) shows that different ‘generations’ of technological artifacts are reflected in the evolution of knowledge trajectories. Within one trajectory, Ethiraj (2007) demonstrates that bottlenecks in the artifact affect the allocation of R&D efforts across the computer industry, and Lee and Berente (2013) use the example of particle filters to show that patenting outside the core component increases once a dominant design for the core component is reached. Lastly, Fontana et al. (2009) briefly mention that the knowledge trajectory of the telecommunication network industry points to an ‘engineering logic’ – which can be interpreted as design hierarchy – governing the sequence of patented inventions, although they do not assess this influence systematically.
However, the evolution of knowledge may differ from the evolution of artifacts in three important respects. First, the body of industry-specific knowledge may exceed what is embodied in commercialized products and services, because firms ‘know more than they make’ (e.g., Brusoni et al., 2001). This means that knowledge generation at any point along the trajectory may not be as focused on specific sub-systems and components as the scope of artifact variation would suggest. Second, firms also make much more than they know, since complex products often employ operating principles that are only imperfectly understood (Vincenti, 1990). Third, not all commercialized knowledge is industry-specific, as firms import a significant share of the knowledge embodied in the artifacts they assemble in the form of components from other sectors (Pavitt, 1984). These last two points mean that some changes on the artifact level may not be reflected in the evolution of the underlying knowledge base. For this reason, processes that depend on the knowledge dimension of technological trajectories, such as knowledge-based competitive advantages of firms and nations (Bekkers and Martinelli, 2012; Epicoco, 2013) or the impact of policy–led incentives on the exploration and exploitation of knowledge (Nemet, 2009; Hoppmann et al., 2013), can only be partially explained using data on the evolution of artifacts. These must be complemented by analyses of the knowledge dimension.
60 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
3. Research Case
3.1. Rationale for Case Selection
For empirical studies of the impact of design hierarchy on the evolution of knowledge, the research case should have three specific characteristics.
First, the product needs to be a systemic artifact with a complex product architecture that has multiple levels in the hierarchy of nested parts and several components on each level, which translates into multiple levels in the hierarchy of control. This allows the possible influence of both types of hierarchy. Second, the product should have been produced for as few applications as possible, ideally with relatively stable demanded service characteristics. On one hand, differences in the demanded service characteristics between applications can lead to the bifurcation of artifact trajectories, making the identification of linkages between knowledge and artifact trajectories difficult. On the other hand, changes in the demanded service characteristics over time can induce changes in the design hierarchy and vice versa (see section 2.3).Yet in order to allow for the observation of their structuring effect on the production of knowledge, both the service characteristics and the design hierarchy should ideally remain unchanged throughout the observed period. Third, the majority of progress over the observed time period needs to have taken place along one technological trajectory, because the phenomenon we want to observe by definition only applies to this type of technological change. Over time, innovative activity along the technological trajectory should ideally have focused on different parts of the system, enabling the sequence of shifts in the focus of inventions to be compared to the sequence of shifts in the focus of innovations.
We selected the case of wind turbine technology in the period 1973-2009 because it fulfills all three requirements, and because understanding the evolution of renewable energy technologies such as wind turbines is particularly important to inform public and environmental policy decisions. After an outline of the scope of the analysis in sub-section 3.2, the three requirements are discussed in detail in 3.3-3.5.
3.2. Scope of Analysis
As it is common in research on dominant designs, we use the concept of a shared operational principle to delineate the scope of our research case (Vincenti, 1990; Murmann and Frenken, 2006). We define wind turbine technology as all technologies pertaining to the conversion of wind energy to electricity
61
by means of a rotor, which is driven by wind and drives an electric generator. This scope includes turbines used for off-grid electricity generation and onshore as well as offshore turbines, but it excludes all wind electricity generators that do not feature a rotor, such as those driven by kites (e.g., as described in patent US 8,319,368). The advantage of applying the shared operational principle to define the scope of analysis is that all included artifacts have a common basic product architecture (Murmann and Frenken, 2006), which allows us to categorize inventions across turbine designs.
3.3. Complex Product Architecture
A typical modern wind turbine consists of a very large number of electrical, mechanical and electronic components which are organized in a complex product architecture, as can be seen in Figure 3 (Section 4.2 describes the derivation of this representation of the product architecture).
Virtually all wind turbine designs feature a product architecture containing the following four groups of components, which we will refer to as sub-systems: (i) a rotor, (ii) a means of converting rotational energy into electrical energy (the power train), (iii) some form of mounting and machine encapsulation (typically the foundation, the tower and the nacelle), and (iv) some form of grid-connection (or electricity storage unit in the case of off-grid generation).10 This common product architecture has multiple levels of nested parts: each of the four main sub-systems contains components, which are made up of sub-components, and so on. The power train (sub-system ii), for example, contains the mechanical drive-train, which contains a gearbox. The gearbox, in turn, consists of cogwheels, shafts, a lubrication system, which are all again made up of various smaller parts. The fact that the product architecture features four sub-systems and three to four components for each sub-system means that the hierarchy of control has multiple levels, too.
10 As an illustration of the complexity of design choices within this common product architecture, Table A1 in the appendix summarizes the main engineering tasks involved in wind turbine design (including the main underlying knowledge domains), and Table A2 illustrates the scope of design decisions for each sub-system and most of the components.
62
Table 1: Product architecture of a typical wind turbine used for grid-connected electricity generation: Figure 3: Components of a wind turbine (adapted from Hau, 2013). Sub-systems and components and their functions in the technological system.
Sub-systems and components Function
Wind turbine (system-level) Conversion of wind energy into grid electricity
Rotor
Blades Conversion of wind energy into rotational energy
Hub Transfer of rotational energy to main shaft
Rotor control system (pitch and yaw mechanisms), control routines Adjustment of rotor and individual blades to wind conditions
Power train
Mechanical drive-train: Rotor shaft, bearings, gearbox, couplings, brake Transmission of rotational energy from rotor blades to generator
Conversion of rotational energy into electrical energy; AC-DC and frequency Electrical drive-train: generatorr, power electronics conversion
Power-train control system and routines Adjustment of drive-train elements to wind & system conditions
Mounting & encapsulation
Nacelle, spinner and bedplate Load carrying; machinery enclosure
Tower Support turbine at designated height; load transfer to foundation
Foundation Load transfer into ground
Climate & vibration control system and routines Regulate operating conditions & minimize system vibrations
Grid connection and/or storage
Transformer / substation and power cables Transfer of electrical energy to grid
Storage (if applicable) Storage of electrical energy
Grid-impact and wind-farm control system and routines Reduce impact of grid-side disturbances; ensure grid-friendly wind farm output
3.4. Stable Service Characteristics
Wind turbines have been produced almost exclusively for onshore, grid-connected electricity generation. Of the roughly 198 gigawatt (GW) installed globally by the end of 2010, only 0.4 GW are small wind turbines (<100 KW), which represent most of the off-grid market (WWEA, 2012), and about 3 GW are installed offshore (GWEC, 2011). This dominance of the onshore, grid-connected market over other segments has prevailed throughout 1973-2009. Therefore, the demanded service characteristics can be approximated as relatively stable in the observed period.
A further benefit for our study arises from the fact that virtually all markets for wind turbines around the globe are created through public policy support. Because the demanded service characteristics are legislated ex-ante, rather than gradually developed by consumers, we can consider them as exogenous to the evolution of artifacts and knowledge. This minimizes the risk of potential endogeneity stemming from the long-term effects of technological change on consumer demand (arrow f in Figure 2).
3.5. Technological Change along one Trajectory
Technological change in wind turbine technology over the last three to four decades has been predominantly characterized by incremental innovations along the trajectory of scaling-up and refining one overarching system design: a horizontal-axis rotor with airfoil-shaped blades that utilize the lift forces of the wind.
Figure 4a shows how the price of wind turbines per watt of electric capacity has come down gradually as the technology progressed along the trajectory. The incremental nature of technological change is also visible in Figure 4b, which shows how the average rotor diameter, turbine capacity and hub height have all increased gradually since 1980.
64 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
a) b) Price of turbines over installed turbine Evolution of turbine size capacity(1984-2011) Tower height / (US data: 1982-2009) Price [$2011/W] rotor diameter [m] Turbine capacity [MW] 4 100 2,0
1984 75 1,5 3 1990 2000 50 1,0
2 2004 25 2011 0,5
1 0 0,0 100 1.000 10.000 100.000 1.000.000 1980 1985 1990 1995 2000 2005 2010 Cumulative installations (log) [MW] Average tower height [m] Germany and Denmark Average rotor diameter [m] Global Average name-plate capacity [MW]
Figure 4: a) Price development of wind turbines over cumulative installations; data from BNEF (2012). b) Size trajectories of modern wind turbines: rotor, name-plate capacity and tower height; data from USGS (2014).
Data on design competition suggests that the focus of innovative activity has shifted over time as the technology has moved along the trajectory (Figure 5a and b). In the late 1970s and early 1980s, firms experimented with different positions of the rotor relative to the tower (upwind, downwind), blade numbers (one, two, three, four or more blades were all introduced commercially) and rotor control mechanisms. As Figure 5a shows, it was not before 1986 that more than 50% of the firms in the market had adopted the three-blade, upwind rotor called the ‘Danish design,’ which is now used in virtually all grid-connected wind turbines.
The focus of innovative activity then shifted within the Danish rotor design toward more efficient power train concepts, as can be seen from the adoption of variable-speed power trains starting from the early 1990s (in white in Figure 5a). The most intense period of design competition on the power- train level was in the late 1990s and early 2000s, when the variable-speed power train with a partial- scale converter emerged as dominant design. It has held more than 50% market share since around 2003.
65
a) b) Evolution of rotor designs (1978-2009) Evolution of power-train designs (1995-2009)
% of firms in the market % market share (in MW) 100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0% 1980 1985 1990 1995 2000 2005 2009 199519971999 2001 2003 2005 2007 2009
Other Fixed-speed power train Light-weight rotors Variable-speed power train w/ variable rotor resistance Danish rotor design (fixed-speed) Variable-speed power train w/ full-scale converter Danish rotor design (variable speed) Variable-speed power train w/ partial-scale converter
Figure 5: Evolution of wind turbine designs: a) the number of firms with different rotor designs in 1978-2009; data from Menzel and Kammer (2011). b) Market share of different power-train designs in 1995-2009; data from Hansen (2012).
What cannot be analyzed with data on artifact evolution such as those presented in Figure 5 are trends in the underlying knowledge base. One can only speculate, for example, whether the surge in variable speed turbines in the 1990s was based on industry-internal refinement in the understanding of wind-specific drive-train requirements or was based on ‘imported’ advances in standardized drive- train components used in other industries. However, these trends directly affect the competitive position of firms and nations, and they have implications for the assessment of innovation policies in the wind industry. Below we proceed to open this black box.
4. Data and Methodology
4.1. Empirical Strategy
In this section, we develop a systematic approach to determining the impact of the design hierarchy on the trajectory of knowledge generation in complex products.
Recent studies of the knowledge dimension of technological trajectories have made significant advances by applying citation-network analysis to patent data (Verspagen, 2007; Fontana et al., 2009; Barberá-Tomás et al., 2011; Martinelli, 2012; Epicoco, 2013). This approach allows researchers to trace the trajectory of knowledge generation over time, but it cannot easily link it to the evolution of
66 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology artifacts and dominant designs (Barberá-Tomás et al., 2011). Studies of the artifact dimension of technological trajectories, on the other hand, have traditionally relied on categorical analysis of product designs available in the market (Rosenkopf and Nerkar, 1999; Frenken and Nuvolari, 2004; Fixson and Park, 2008; Mendonça, 2012). This approach is useful for analyzing the influence of the design hierarchy on the evolution of artifacts but does not allow identification of developments in the underlying knowledge base. Combining these two approaches allows us to identify the influence of the design hierarchy on the trajectory of knowledge generation and thus to bridge the knowledge and artifact dimensions of technological trajectories.
Our empirical strategy was as follows: We first used a combination of desk research and expert interviews to identify the product architecture, relevant service characteristics and design hierarchy of wind turbines (Section 4.2). Second, we analyzed connectivity measures in the network formed by wind turbine patents and patent citations in order to identify the core trajectory of knowledge generation (the data is described in Section 4.3 and the algorithms in 4.4). In particular, we analyzed how the core trajectory of knowledge generation evolved and converged over time, by investigating how the core of the knowledge trajectory changes as an ever-increasing number of patents add to the knowledge base over time. We then analyzed in detail the core trajectory of the complete knowledge base, which we will refer to as ‘today’s core trajectory,’ to determine how the focus of inventive activity along the trajectory shifted over the course of the last four decades. In a third and final step, we manually categorized the core patents on the trajectory of knowledge generation, identified in step two, according to their focus in the design hierarchy (Section 4.5). Taken together, these three steps yield a unique database of key inventions along the trajectory of knowledge generation that allows us to trace how the trajectory gradually proceeds through the wind turbines’ design hierarchy.
4.2. Design Hierarchy
The design hierarchy was identified through a qualitative assessment of the product architecture, the relevant service characteristics, and the linkages between the two.
We first developed an initial understanding of the product architecture from the technical literature. Then this initial understanding was iteratively refined through five semi-structured telephone interviews with two industry professionals. The resulting product architecture is shown in Figure 6.
67
The list of relevant service characteristics was identified through a series of nine structured interviews,11 in which we asked for characteristics that determine model choice. From the resulting long list of criteria we removed turbine model-specific characteristics such as the availability of upgrades and spare parts as well as purely organizational characteristics such as warranty time, contract flexibility, reaction time, etc. We further aggregated some criteria to reduce complexity. (The final selection is shown in the column headers of Table A4 in the Appendix.)
Lastly, the design hierarchy, which is determined by the linkages between the product architecture and the service characteristics, was developed through structured interviews with two industry professionals, in which we asked them to link sub-systems and components of a wind turbine to the identified list of service characteristics. We contacted the interviewees a second time to clarify inconsistencies between the two and removed linkages where disagreement could not be resolved.
4.3. Patent and Patent Citation Data
We used patents as indicators of knowledge generation in the wind industry (Nemet, 2009) and citations as indicators of technological relatedness (von Wartburg et al., 2005).
For the underlying patent database, we compiled wind patents12 filed from 1963 to 2009 from the Derwent World Patent Index (DWPI) database, which contains patents from 48 patent-issuing authorities worldwide.13 We chose DWPI because it facilitated the assessment of patent content by providing expert-generated abstracts of all patents (see Section 4.5), including translated abstracts for non-English entries in the database. The search was conducted in early 2013 in order to account for the time-lag between patent filing and publication of patents filed in 2009.
The patent database was compiled by applying a list of keywords to the titles, abstracts and claims of patents in 20 four-digit International Patent Classification (IPC) classes. We extracted an initial list
11 For this step we interviewed professionals (by telephone and on-site) from two wind turbine operators and wind turbine experts from one insurance company, one engineering service provider, one bank, one consultant and one project developer. 12 We used patent families as the unit of analysis to avoid double-counting of multiple filings. 13 Even though our focus is on the time period 1973-2009, the database includes patents from 1963-1972 in order to improve the results of the connectivity analysis for the earliest patents.
68 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology of relevant keywords from the technical literature (four industry experts provided feedback on the identified keywords) and an initial set of wind-related IPC classes from the ‘Green Inventory’ of the World Intellectual Property Organization. We then iteratively curtailed the keyword list and IPC classes by manually checking random samples of patents for irrelevant keywords, and we added further IPC classes by analyzing in which IPC classes relevant patents in the database were co-filed. The combination of keywords and IPC classes yielded a total of 25,512 patent families (including applications and granted patents). After retrieving the citation data of all patents, we extended the database in a second iteration to include those 1,000 outside patents that received the most citations from the patents in the database (almost all of these are wind patents). Tests indicate the presence of about 6% false positives and 9% false negatives in the final dataset.14
The citation data was extracted from the DWPI and in addition from the Thompson Innovation database. Neither of the two databases alone provides citation data for the full period from all patent offices that we deemed important for the case of wind power, but taken together the coverage is satisfying.15 We cleaned the citations of duplicates and excluded all patents that were not connected to other patents in the network. Finally, we reversed the citations to transform them into indicators of knowledge inheritance between nodes in the network, and we excluded circular references 16 (Martinelli and Nomaler, 2014). The final database contains 11,330 patent families with 41,268 citations between them (network A in Table 2).
14 To test for false positives, we randomly tested a total of 1,000 patents (50 patents for each of the 20 four-digit IPC classes in the search string). For false negatives, we checked how many of the patents filed by the top 8 pure-player wind turbine manufacturers (in 2010 by market share) were included in our database. 15 We considered as important the 12 countries with the most successful turbine manufacturers (by market share) in the observed period as well as the multilateral patent offices (in country codes of the World Intellectual Property Organization): BE, CN, DK, DE, ES, EP, GB, IN, IT, JP, KR, NL, US, WO. Gaps that remained even after combining citation data from both databases are: BE before 1987, CN, ES before 1992, IN, IT before 1986, KR before 2008. 16 Whenever we found circular references, i.e., mutual citations between patents, we deleted the citation coming from the patent with the earlier priority date. Such citations can occur when examiners add citations to new patents filed during the examination process, or when patents are filed in multiple countries.
69
4.4. Patent-Citation Network Analysis
Connectivity analysis of networks created from patents (as vertices) and patent citations (as arcs) has emerged as a standard approach for analyzing the knowledge dimension of technological trajectories (Mina et al., 2007; Fontana et al., 2009; Barberá-Tomás et al., 2011; Bekkers and Martinelli, 2012; Martinelli, 2012; Epicoco et al., 2014). We employed connectivity analysis with two objectives: to investigate how the core trajectory of knowledge production evolved over time, and to analyze the foundations of today’s core trajectory in detail to determine how the focus of inventive activity along this trajectory shifted over the course of the last four decades. For both objectives, we used connectivity algorithms to extract sub-networks that could then be categorized manually (see Section 4.5).
To address the first objective, we extracted a series of gradually growing sub-networks that allowed us to analyze how the core trajectory of knowledge generation in the wind industry varied and converged over time. This approach reflects the fact that the core trajectory of knowledge generation identified in an industry’s knowledge base at any point in time, represented here by a set of patents, changes ex- post when new knowledge is added over time: the (patented) roots of what the industry is working on today may have been outside the industry’s focus of knowledge generation at the time they were filed, and patents inside the focus in the 1980s may have become obsolete by now.
We began by specifying a series of gradually growing networks Nt, in which each Nt contains all patents filed between 1963 and the year t=1975…2009 and the citations between them (network set B in Table 2).17 We only included citations with a lag between the application dates of the citing and cited patents of no more than five years so as not to disproportionately weigh older patents that had more time to get cited. For each Nt we applied the search path link count (SPLC) algorithm (Hummon and Doreian, 1989; Verspagen, 2007). This allowed us to determine vertex and arc weights, which represent the importance of patents and citation linkages for the cumulative evolution of technological knowledge represented by the network, and act as input to the connectivity algorithms described below.
17 The year 1975 was chosen as a starting point because at that time the cumulative number of patents exceeded 100.
70 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
We then used the critical path method to identify the ‘backbone’ of each network Nt, which can be understood as a core trajectory of knowledge generation in the observed period (Mina et al., 2007; Fontana et al., 2009; Barberá-Tomás et al., 2011; Bekkers and Martinelli, 2012; Martinelli, 2012; Epicoco et al., 2014). Thereafter, we extracted each resulting critical path as a separate sub-network – one for each Nt (network set C in Table 2) – and categorized all contained patents according to their content (see Section 4.5). By displaying the sub-networks individually and identifying change and stability over time, we were able to observe how the core trajectory of knowledge evolution varied and converged over time.
To address the second objective, investigating in detail the focus of inventive activity along today’s core trajectory over the last four decades, we started with the full network (1963-2009) and again used the SPLC algorithm to weigh vertices and arcs. Instead of using the critical path method, however, we extracted the two sub-networks containing 80% and 95% of the total vertex weight, respectively (networks D and E in Table 2). Because the weight of patents in the network is highly skewed, with a few patents holding most of the aggregate weight, this vertex-cut algorithm (Batagelj and Mrvar, 2004) reduces the number of patents in the network significantly – in our case from 8,907 to 494 for 95% of the aggregate vertex weight and 158 for 80% (see Table 2). This allows us to approximate characteristics of the full network, such as the focus of inventive activity in the design hierarchy, by categorizing only a relatively small subsample. In particular, we can obtain a close approximation of the weighted average of the focus of inventive activity in the full network of 8,907 patents by manually categorizing only 494 patents.
71
Table 2: Descriptive statistics of (sub-) networks and their role in the analysis.
Patents (Sets of) Number of Manually Content Time period (citation Analysis steps Networks networks coded (y/n) links)
11,330 A All patents 1 1963-2009 No --- (41,268)
Sequential full 111 1963-1975 … networks, (43) … Calculation of vertex and arc weights B 35 1963-2009 No citation links ≤5 8,907 to determine critical paths (see set C) (in year-steps) years (18,718)
1963-1975 … Sequential 4 (3) … Variation of core trajectory over time C 35 1963-2009 Yes critical paths 33 (32) (Figure 7) (in year-steps)
Patents with top Analysis of focus of inventive activity D 80% of vertex 1 1963-2009 158 (499) Yes along today’s core trajectory (Figure 8) weight
Patents with top 494 Robustness check for analysis of E 95% vertex 1 1963-2009 Yes (1,827) dominant knowledge trajectory weight
Patents with top Analysis of knowledge flows between 80% vertex F 1 1963-2009 158 (817) Yes patents on today’s core trajectory weight (Figure 8) (all citations)
4.5. Patent-Content Analysis
As a final step, we manually coded the abstracts and claims of the patents in the sub-networks extracted in Section 4.4 to identify how the industry’s knowledge base evolved over time (networks C- F in Table 2). One mechanical engineer and one electrical engineer independently coded each of the patents according to the abstracts’ focus and located them in the design hierarchy.
The coding scheme we used in the analysis, shown in Table 2, has three levels in the hierarchy of nested parts (system, sub-system and component) and four levels in the hierarchy of control on the sub-system level (rotor, power train, mounting & encapsulation and grid connection). 18 The
18 The initial coding scheme also had a sub-component level. However, the agreement between the two coders was not high enough to justify a distinction between the component and sub-component level (<70%), and in all but one component the agreement between the two coders on the distinction between different sub-components (such as between generators and power electronics) was also insufficient (<80%).
72 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology agreement between the two coders was 89% in the hierarchy of nested parts and 92% in the hierarchy of control.
We cross-checked the resulting focuses of knowledge generation along the trajectory in a final round of interviews with four academic experts on the wind industry. All four confirmed the trends displayed in the data.
Table 3: Coding scheme for patent focus.
Content code Content Example
Vertical axis turbine with novel rotor and Novel wind-turbine design in which novelty has to do with the design of Wind turbine novel drive-train arrangement (US 3,902,072) at least two sub-systems (rotor, power train, mounting & encapsulation, (system-level) or horizontal-axis rotor with rotor-integrated and/or grid connection) generator (US 4,289,970)
Rotor Novel rotor design in which novelty has to do with the design of at least Rotor arrangement with teetering hub and (sub-system level) two components (blades, hub and/or rotor control) rotor control mechanism (US 4,201,514)
Rotor Novel rotor design in which novelty has to do with the design of one Sectioned rotor blade (US 4,389,162) (component level) component (blades, hub and/or rotor control)
Novel power-train design in which novelty has to do with the design of at Power train least two components (mechanical transmission system, generator, power Compact, gearless power train (US 6,921,243) (sub-system) electronics, power-train control)
Novel power train design in which novelty has to do with the design of Power train one component (mechanical transmission system, generator, power Planetary gearbox (US 6,420,808) (component) electronics, power-train control)C
Mounting & Novel mounting & encapsulation design n which novelty has to do with Novel tower-nacelle arrangement in which encapsulation the design of at least two components (nacelle, spinner, bedplate, tower, transformer is mounted inside the top of the (sub-system) foundation, climate & system-vibration control) tower (US 7,119,453)
Mounting & Novel mounting & encapsulation design in which novelty has to do with Tower consisting of pre-fabricated modules encapsulation the design of one component (nacelle, spinner, bedplate, tower, (US 7,770,343) (component) foundation, climate & system-vibration control)
Novel electrical connection of wind turbines in Novel grid-connection design in which novelty has to do with the design Grid connection a wind farm, including substation and of at least two components (mechanical transmission system, generator, (sub-system) individual transformers and cabling (US power electronics) 7,071,579)
Novel power train design in which novelty has to do with the design of Control system for wind farm that optimizes Grid connection one component (transformer, substation, cabling, storage, wind-farm voltage and reactive power output (US (component) integration control, grid-fault control) 7,119,452)
We were further able to test the robustness of the coding by assessing whether or not the categorization of sub-systems is reflected in the citation data, because previous research has shown that patent citations are more likely to link patents within than across sub-systems and components (Rosenkopf and Nerkar, 1999). χ2 tests for the randomness of the distribution of citations from each
73
of the four sub-systems indicate that the results of the coding do indeed correspond to relational patterns in the citation data (see Table A3).
5. Results
5.1. Design Hierarchy
The design hierarchy, displayed in Figure 6, is derived from the interplay of the product architecture and the service characteristics. The product architecture directly yields the hierarchy of nested parts, with the turbine system on the system-level, the rotor, power train, mounting & encapsulation and grid connection on the sub-system level, and all other elements on the component level.
The hierarchy of control is determined on the sub-system level by assessing the influence of each sub- system on the service characteristics. Specifically, each sub-system’s position in the hierarchy of control is calculated from the number of service characteristics affected by the sub-system. The underlying relationships between system elements and service characteristics are presented in Table A4 n the Appendix (Murmann and Frenken, 2006). Our results suggest that the hierarchy of control of a wind turbine follows the order (from core to periphery) (i) rotor, (ii) power train, (iii) mounting & encapsulation and (iv) grid connection, as indicated by the vertical order of the sub-systems in Figure 6.19
19 The resulting design hierarchy is in line with the prominent role that rotor and power-train designs assume in historical accounts of wind turbine engineering (Karnøe, 1993; Gipe, 1995; Garrad, 2012).
74 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Hierarchy of nested parts
Wind turbine system
Rotor Components (blades; hub; rotor control system: pitch and yaw) Power train Components (mechanical drive-train: rotor shaft, bearings, gearbox, Hierarchy of couplings; electrical drive-train: generator, power electronics; power-train control control system and routines
Mounting & encapsulation Components (nacelle, spinner and bedplate; tower; foundation; climate & vibration control system and routines Grid connection Components (transformer / substation and power cables; storage; grid- impact and wind-farm control system and routines
Figure 6: Design hierarchy of a wind turbine, based on product architecture presented in Table 1 and pleiotropy map presented in Table A4 in the appendix.
5.2. Gradual Stabilization of Knowledge Trajectory
The evolution of the core trajectory over the last four decades was analyzed iteratively by determining the core trajectory as increasingly more years of data are added to the patent-citation network. Figures 7a-h shows how the core trajectory meanders through the design hierarchy for eight networks representing network growth in 5-year steps. It can be seen how the knowledge trajectory in the industry varies substantially from 1974, the year when the earliest core trajectory begins, until 1990, but stabilizes thereafter. This result is quantified in Figure 7i, which displays for each year t the percentage of the patents on the core trajectory of network Nt (i.e., the network with data until year t) that are no longer on the core trajectory of Nt+5. Only by 1990 does this hazard rate, which is a measure of variation of the core trajectory, remain consistently below 50% (the value in 1989 is exactly 50%). Accordingly, our analysis is able to describe the competition between fundamentally different engineering approaches in the 1970s and 1980s as well as the subsequent convergence on the ‘Danish’ bottom-up approach to wind turbine design. This convergence had been described extensively, but only in qualitative studies (e.g., Karnøe, 1993b; Gipe, 1995; Johnson and Jacobsson, 2000; Garud and Karnøe, 2003; Nielsen, 2010). The sequence of core paths in Figure 7 indicates that the knowledge trajectory stabilized as soon as the core patents on the rotor stabilized: while there is much variation between the rotor-level patents in the networks N1975 -N1990, there is no further change on the rotor level from N1995 on, which coincides with the stabilization of the knowledge trajectory overall. This
75
suggests that the dominant rotor design reduced variation on the highest level in the hierarchy of control, but set the agenda for further developments and thus allowed for much innovation on lower levels of the design hierarchy (cf. Clark, 1985; Sahal, 1985; Frenken, 2006; Murmann and Frenken, 2006).20
a) 1963-1975 b) 1963-1980 c) 1963-1985 1975 1985 1995 2005 1975 1985 1995 2005 1975 1985 1995 2005 Turbine (system-level)
Rotor
Power train
Mounting & encapsulation
Grid connection
d) 1963-1990 e) 1963-1995 f) 1963-2000 1975 1985 1995 2005 1975 1985 1995 2005 1975 1985 1995 2005 Turbine (system-level)
Rotor
Power train
Mounting & encapsulation
Grid connection
Share of patents no longer on critical path 5 years later (%) g) 1963-2005 h) 1963-2009 i) 1975 1985 1995 2005 1975 1985 1995 2005 -1975 -1985 -1995 -2005 Turbine (system-level) 100
Rotor 80 60 Power train 40 Mounting & encapsulation 20
Grid connection 0
Legend Designs
Patents (size ~ network weight) Vertical axis Horizontal axis, upwind, three-bladed Citation (reversed) Horizontal axis, downwind Horizontal axis, upwind, three-bladed, variable speed, direct-drive Horizontal axis, upwind Horizontal axis, upwind, three-bladed, variable speed, gearbox
Figure 7: Variation of core trajectory over time. a)-h) display the critical paths in a series of gradually growing networks (time periods are given in parentheses). Figure i) shows the five-year hazard rate for patents on the core path, indicating how the core trajectory gradually stabilizes over time. The size of vertices and arcs represents their weight. The shade of the vertices indicates the rotor and power-train design underlying the invention; patents without any specific rotor and/or power-train design are colored as black.
20 Table A5 in the appendix provides details on content and assignees of the patents along the top path of the core trajectory.
76 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
5.3. Foundations of Today’s Trajectory of Knowledge Generation
The analysis of the networks with patents that represent 80% and 95% of the vertex weight in the network confirms the overall trend shown in Figure 7h, while adding further depth and detail. Figure 8 shows the network of those patents that account for 80% of the vertex weight. Significant inventions along the trajectory can be found in all four sub-systems and across all levels of the hierarchy of nested parts, underlining the complex, systemic character of wind turbines. However, the focus of inventive activity shifted through the system in a clearly sequential way: from the rotor to the power train, grid connection and lastly mounting & encapsulation.21
1975 1980 1985 1990 1995 2000 2005 2010
Turbine (system-level)
Rotor Sub-system level Component level
Power train Sub-system Component
Mounting & encapsulation Sub-system Component
Grid connection Sub-system Component
Legend Citation ≤ 5 year lag (width ~ link weight) Patent (size ~ node weight) Citation > 5 year lag
Figure 8: History and foundations of today’s trajectory of knowledge evolution in the wind industry in detail (represented here by the network containing 80% of the total vertex weight). Citations with lag >5 years were not used in analysis, but are displayed here to indicate actual technological linkages.
An analysis of the 95%-weight network, shown in Figure 9, provides further quantitative evidence for (i) the highly sequential pattern of knowledge generation along the trajectory, and (ii) the structuring effect of the hierarchy of control on the underlying sequence. Expressed in terms of the hierarchy of nested parts, across the observed period, most inventive activity is on the component level (Figure 9a),
21 Notably, the full sequence through all sub-systems took more than 30 years (from 1975 to around 2005, when the last sub-system was reached), highlighting the complex, systemic nature of wind turbines.
77
while there is no clear trend in the inventions on the system- and sub-system levels. In contrast, the hierarchy of control is well reflected in the sequence of inventive activity along the trajectory (see Figure 9b and Table A6 in the Appendix): Rotor patents account for 77% of the total vertex weight in 1975-1979 and for 76% in 1980-1984, but only for an average of 3% in the two subsequent decades. Power train patents, on the other hand, surge from 16% in 1980-1984 to 91%, 87% and 78% in the three periods after that, before falling to 34% in 2000-2004. The last two periods are dominated by grid connection patents, with 46% in 2000-2004, and mounting & encapsulation with 32% in 2005- 2009, two categories that both had 0 patents in the 95%-weight network in 1985-1989.22 Toward the end of the observed period, the focus of knowledge generation seems to diffuse.23
a) b) Hierarchy of nested parts over time Hierarchy of control over time (sub-system level) % of patents (weighted) % of patents (weighted) 100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0% 1975 1980 19851990 1995 2000 2005 2009 1975 1980 1985 1990 1995 2000 2005 2009
System level Rotor Sub-system level Power-train Component level Mounting & encapsulation Grid connection
Figure 9: Weighted-average of the focus of patents over time: (a) hierarchy of nested parts and (b) hierarchy of control; data from patents representing 95% of vertex weight in the network (determined through a search path link count algorithm). Data displayed as 5-year moving averages.
The presented results suggest that the design hierarchy had a structuring effect on the trajectory of knowledge evolution in the wind industry, albeit with two qualifications. First, although the earliest
22 An ordered, bivariate ordinal regression of the hierarchy of control on the logarithmized cumulative number of patents in the network confirms that inventive activity gradually shifts downwards on the hierarchy as the knowledge base grows ( =0.60; t(494)=6.53; p<0.001, AIC=1437). 23 This observation might be partly due to the fact that patents had lower chances of being cited in the last four years.
78 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology patents in the field of mounting & encapsulation precede those in grid connections, significant activity in the latter field occurred earlier (after 1995). This coincides with the first regulations on grid-compatibility in the industry in the late 1990s, which suggests that the early shift was due to the temporarily heightened urgency of a single service characteristic (grid compatibility; discussed further in section 6.4). Second, the hierarchy of nested parts appears not to be a good predictor of the sequence of knowledge generation along today’s knowledge trajectory. On one hand, inventive activity did not start on the system level (at least not in patents filed from 1963).24 On the other, in all four subsystems the earliest inventive activity is on the component level, rather than on the sub-system level. And in all sub-systems but the rotor, which features significant patents on the sub-system level early on, the vertices on the component level appear much more important than those one level higher. One possible explanation for this second qualification is that inventors had network-external knowledge on the system-level and sub-system level to build on, for example in the form of standardized generators, towers and transformers available in other industries, and thus could focus immediately on wind-specific improvements on the component-level. To shed more light on this possible explanation, we analyzed below the relative importance of network-external knowledge over time.
5.4. Influence of Network-External Knowledge along Trajectory
Figure 10 plots the influence of external knowledge, measured by the weighted-average share of citations to patents outside the network. As can be seen in Figure 10a, which shows the relationship for the full 95%-weight network, the influx of industry-external knowledge consistently declines over time. However, as indicated by Figure 10b, which shows the relationship for each sub-system separately, this decline is not uniform across the elements of the system; rather, each sub-system features very different rates of decline.
The data is plotted over the cumulative vertex weight, rather than over time, to show that the decrease is primarily a function of how much knowledge has been generated within the specific sub-system,
24 However, system-level inventions did have some impact later. The patents on the system-level in the late 1990s as well as the recipient patents on lower levels relate to direct-drive technology, a specific type of power train that does not need a gearbox.
79
rather than along the trajectory in general. This means that external knowledge gained importance whenever the focus of inventive activity shifted to a new sub-system.
The high intercept of the rotor function indicates that very little network-internal knowledge was available to build upon, but the relatively steep slope suggests that it was built up as a largely independent knowledge base. In contrast, the lower intercept of the power-train, mounting & encapsulation, and grid-connection functions reflects their being in focus in later stages of the trajectory: inventions in these sub-systems could partially build on a wind-specific knowledge base that had already been developed. However, their more moderate slope also indicates that there was more persistent import of external knowledge. This reflects the fact that, unlike rotor blades, these sub-systems and components have much in common with other electro-mechanical machinery.
a) b) Impact of outside knowledge over time: 95%- Impact of outside knowledge over time: four weight network different sub-systems Share of citations to Share of citations to outside patents outside patents 1.0 1.0 1975-79 Rotor 0.8 0.8 Power train 80-84 90-94 Mounting 0.6 0.6 & encapsulation 85-89 Grid connection 95-99 0.4 0.4 Power train 05-09 2000-04 0.2 0.2
Grid connection Rotor M&E 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.1 0.2 0.3 0.4 0.5 Cumulative vertex weight in network Cumulative vertex weight in sub-system Figure 10: Declining impact of outside knowledge over time: share of citations to patents outside the network over cumulative weight. Figure a) shows the relationship for full 95%-weight network; b) shows the relationship for each of the four sub-systems separately.
6. Discussion
6.1. Creative Sequences in the Evolution of an Industry’s Knowledge Base
Our results help explaining how the knowledge base of a complex product emerges and grows over time. In particular, this paper provides a model that explains how the focus of inventive activity shifts along the technological trajectory through sub-systems and components of the product, and how the impact of external knowledge evolves over time along with the shifting focus. This model holds that
80 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology the evolution of an industry’s knowledge base along a technological trajectory is a creative sequence, with sequential changes in the focus of inventive activity and cyclical changes in the importance of industry-external knowledge, rather than a more or less linear process of incremental growth and refinement.
The principal finding of our paper is that the focus of knowledge generation shifts in a highly sequential way through the clusters of technological problems that pertain to different sub-systems of a complex product. The order underlying this creative sequence is strongly influenced by the core- periphery dimension of the design hierarchy: Our findings suggest that if a systemic artifact has many different sub-systems, inventive activity will focus first on the (core) sub-systems that are most important for the demanded service characteristics. The knowledge trajectory in the industry will stabilize once the understanding of the design of this sub-system has reached some degree of saturation. It will then gradually proceed, along the sequence defined by the design hierarchy, toward more peripheral sub-systems.25 This pattern means that the design hierarchy defines not only the physical interaction of sub-systems and components in the artifact, but also structures the sequence in which the knowledge base of the industry is expanded and refined in different directions.
Our second finding is the recurring influence of outside knowledge along the creative sequence, which explains why the nested-parts dimension of the design hierarchy appears to have no influence on the trajectory of knowledge generation. Every time the focus of knowledge generation shifts to a new sub- system, a new wave of integration of network-external knowledge is initiated, starting with a high influence of network-external knowledge sourcing that gradually declines as the industry builds an independent understanding of the sub-system in focus. A deeper look at the sources of knowledge of the inventions on the trajectory suggests that the industry built upon two sources of network-external knowledge on the system and sub-system levels: industry-internal knowledge that pre-dates our observation period and industry-external knowledge. By drawing from these sources of industry- external knowledge, knowledge generation could skip levels in the hierarchy of nested parts.
The first source, industry-internal knowledge that pre-dates our analysis (even though our observation period covered roughly 50 years), explains the lack of system-level patents on today’s core trajectory.
25 Our data did not allow us to analyze the trajectory on the component level, but we would expect a similar pattern there.
81
Due to the necessarily limited time period that our database covers, the fundamental system design of horizontal axis, lift-based wind turbines, was well-established at the beginning of the observed period (even though its application to large-scale electricity generation was a novelty in the universe of artifacts). The fact that our database begins in 1963 means that system-level inventions such as US2037528 (filed in 1934) or US 2622686 (1948) cannot be located on the trajectory.
The second source, industry-external knowledge, explains the lack of patents on the sub-system level before patenting begins on the component level. The knowledge base of the wind industry builds on knowledge transferred from a number of adjacent sectors, including aerospace, electrical engineering, ship building and agricultural machinery (a list of the main involved knowledge domains is given in Table A1). Knowledge from these adjacent sectors entered the wind industry in the beginning in the form of sub-system assemblies – the power-train of a wind turbine is in principle not much different from that of a hydro turbine – as well as standardized components such as gearboxes, generators and towers. The adoption of these components in the wind industry meant an innovation in the universe of artifacts, but not necessarily novelty in the evolution of knowledge. When the focus of inventive activity later shifted toward these components (e.g., to the power train in the late 1980s), the generation of wind-specific knowledge did not start with the sub-system level, but with specific adaptations of standard components to the operational requirements of a wind turbine. Indeed, on each sub-system, the earliest patents on the core trajectory are component-level inventions that – in addition to wind-turbine patents – draw significantly on conceptual patents from other sectors. For example, MAN’s rotor patent US 4,297,076 cites water wheels (such as US 2,152,984), United Technologies’ power-train patent US 4,297,076 builds on technology from aircraft engines (US 4,330,743) and ABB’s grid-connection patent US 6,670,721 references many generic grid-related patents (such as US 6,429,546).
6.2. Implications for Technology Strategy and Public Policy
Our model of creative sequences has implications for technology strategy and public policy aimed at stimulating innovation in complex products. Both derive from the sequential pattern of knowledge creation and the influence of the design hierarchy.
The focus of an industry’s inventive activity directly affects the competitive value of knowledge held by firms and nations. Our model of creative sequences suggests that movement along the trajectory does not preclude dramatic shifts in the value of knowledge positions of firms and nations. On one
82 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology hand, at any point in time, the knowledge – codified in patents – that has a long-lasting impact on the trajectory of knowledge generation focuses on only a very narrow set of technological problems. On the other hand, this narrow focus shifts over time between sub-systems, which may depend on entirely different knowledge bases. For example, while the rotor of a wind turbine requires understanding of structural engineering, aerodynamics and materials, the power train requires knowledge of electrical engineering and electronics. This pattern may help explain sudden shifts in an industry’s competitive landscape that occur without major shifts of the technological trajectory, such as the sudden rise of large electrical engineering conglomerates in the wind industry in the 2000s (including GE, Siemens, Alstom and Areva) that coincided with a shift in the focus of inventive activity toward power-train control and grid integration issues.
The notion of creative sequences also has implications for technology policy. Many governments are attempting to steer technological change in complex products to improve the competitive and environmental performance of high-technology sectors. In recent years, a particular focus has been placed on policies aimed at increasing demand for specific innovative products, such as direct financial incentives for solar PV or wind power. In the academic debate, an argument against such subsidies has been that market creation for emerging technologies predominantly leads to the exploitation and refinement of the existing knowledge base, rather than the exploration of new and potentially more radical solutions, and that this may not be enough to achieve long-term policy goals (Menanteau, 2000; Sandén and Azar, 2005; Nemet, 2009; Hoppmann et al., 2013). Our results suggest a more nuanced understanding: movement along the trajectory does not preclude the exploration of novel solutions, based on industry-external knowledge, on the sub-system and component levels. The development of direct-drive power trains on the sub-system level (power train) is a good example of this: although they constitute a development along the trajectory, direct-drive power trains involved the integration of industry-external knowledge of permanent magnets and full-scale power converters, and facilitated a step-change in performance (especially in terms of grid behavior). Numerous other historical examples, which include jet engines in airplanes, automatic transmissions in automobiles, the computer mouse and random access memory, also indicate that sub-system level innovations can drive major system-performance improvements. This means that if the system is sufficiently complex, movement along the trajectory driven by policy-induced demand may thus well lead to significant external knowledge sourcing and exploration of new solutions.
83
6.3. Interaction of Artifact and Knowledge Dimensions along Technological Trajectories
Our extension of the methodology introduced by Verspagen (2007) and others will allow researchers to study the knowledge and the artifact dimensions of technological trajectories in an integrated way. We believe that the presented methodology can yield particularly valuable insights in two directions.
First, it can be used to study the interaction between the knowledge and artifact dimensions of technological trajectories in greater detail. If data on the knowledge trajectory is systematically compared to data on product designs and market shares, further conclusions may be drawn about the mechanisms of influence between the two. In particular, future research could investigate the relative timing of shifts in the knowledge trajectory and the emergence of dominant sub-system designs in the market. Our results for the case of wind turbines suggest that different modes of innovation were prevalent in different parts of the trajectory. Interestingly, there is variation on the rotor-level of the core trajectory until 1991, while the dominant rotor design in the market (>50% from 1986) had emerged about five years earlier (cf. Figure 5 and Figure 7). This points to a non-linear model of innovation in the design of wind-turbine rotors and an important role of learning by doing and using in the early years of the industry. The shift away from the power-train level (around 1997), however, took place long before the dominant design had been established in the market (>50% from 2003). This indicates a more highly linear model of innovation in this period. The shift from a non-linear to a more linear relationship between knowledge production and artifact commercialization in the 1990s corresponds well with qualitative accounts of the wind industry (Garud and Karnøe, 2003; Hendry and Harborne, 2011; Garrad, 2012). This means that differences in learning mechanisms can be observed when comparing the evolution of knowledge and the evolution of artifacts. It also means that shifts in the predominant mode of innovation might be rooted not only in the maturation of the industry, but also in differences in the technological nature of the two sub-systems, in this case the rotor and the power train.
Second, our results suggest that the methodology can be used as a meaningful proxy for the evolution of artifacts along the hierarchy of control. In many cases this can facilitate a deeper look into a technological trajectory’s inner dynamics, since many technological developments may be concealed when only data on design specifications in the market is examined. Patent data is relatively easy to access and process, whereas data on commercialized designs may not always be available in standardized form and sufficient detail. For example, our analysis allowed us to analyze how
84 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology knowledge generation shifted across intangible components such as wind-farm integration strategies and power train control systems. 26 Our results also point toward the ability to approximate the emergence of a dominant design in a specific component that cannot be easily observed statistically by analyzing the shift of the knowledge trajectory away from that component. Furthermore, the richness of patent data may facilitate detailed analyses of the role of different types of actors along the trajectory, as well as spatial aspects of technological evolution.
6.4. Limitations and Future Research
Two assumptions that limit the generalizability of our findings are worth noting. First, we assumed that the hierarchy of design decisions is stable over time and across countries. This assumption could be relaxed for a more detailed analysis of specific regions or time periods. On one hand, service characteristics may not always be equally important, and their weighting may depend on characteristics of customers, institutions and geographies. On the other hand, service characteristics and their weight may change over time as customers learn about technology and their needs. These limitations offer fruitful avenues for future research. Second, in identifying the trajectory of knowledge generation, we approximated it with patented inventions. This introduces a bias against knowledge that is openly shared, tacit or protected through means other than patenting. In the case of wind turbines, the knowledge pertaining to blade production in particular is typically not patented but protected as a trade secret. The fact that we found very few process patents along the trajectory may be due to a bias against process knowledge in general. Furthermore, many small wind turbine manufacturers did not patent much in the early years of the industry, possibly causing our analysis of the variation of core trajectories over time (Figure 7) to underestimate how early the industry converged on today’s core knowledge trajectory.27 Future research could apply qualitative methodology to capture the evolution of knowledge more holistically along the trajectory.
26 Patent-citation data may also serve to identify the product architecture itself, in a methodology similar to that of Baldwin et al. (2014). 27 Our analysis of the foundations of today’s core trajectory, shown in Figure 8, should be unaffected by this bias because the algorithm identifies the foundations and history of today’s core trajectory ex-post.
85
7. Conclusion
Studies of technological evolution provide ample evidence that a product’s hierarchy of design decisions, or design hierarchy, influences the evolution of artifact designs available in the market. Much less is known about the design hierarchy’s effect on the evolution of the knowledge base of an industry. For such an analysis, this paper employed the case of wind turbine technology over the period 1973- 2009. We developed a methodology by linking a recently developed, quantitative approach to studying the knowledge dimension of technological trajectories to methods for studying the evolution of systemic artifacts. This novel approach allows us to relate systemic relationships between sub- systems and components in the physical artifact to patterns in the evolution of knowledge, and it may facilitate a better understanding of the interaction between the knowledge and the artifact dimensions of technological trajectories.
Our results unmask a sequential pattern in the emergence of an industry-specific knowledge base along the technological trajectory, structured by the product’s design hierarchy: The trajectory of knowledge generation is marked by creative sequences, the focus of which shifts over time between different sub-systems, with each shift initiating a new cycle of integration of industry-external knowledge into the knowledge base.
These findings have implications for the literature on knowledge positions and competitive advantage. Whenever sub-systems of an artifact depend on different knowledge domains, windows of competitive opportunity for firms and nations with knowledge in adjacent sectors can arise along the trajectory, if the adjacent sector is related to the sub-system that moves into the focus of innovation. In other words, what constitutes a good knowledge position to enter a specific industry may change significantly over time. This may help explain – and even anticipate – shifts in the competitive landscape that occur in the absence of discontinuities in the trajectory.
Our findings also have implications for the literature on the innovation impact of demand-side policies. The pattern of creative sequences implies that public policy-driven incentives that induce movement along a technological trajectory – rather than stimulating new trajectories – may not only induce the exploitation and refinement of existing knowledge, but also induce the exploration of new knowledge and concepts on the sub-system and component levels of the design hierarchy.
86 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Acknowledgements
The authors would like to thank the participants of the Science, Technology and Public Policy seminar at Harvard in November 2013, the International Sustainability Transitions conference at ETH Zurich in June 2013 and the Energy Systems in Transition conference at Karlsruhe Institute of Technology in October 2013, as well as Catharina Bening, Etienne Eigle, David Goldblatt, David Grosspietsch, Joern Hoppmann, Annegret Stefan, Stephan Stollenwerk and Kavita Surana for their valuable inputs. All errors remain our own.
87
References
Abernathy, W.J., Clark, K.B., 1985. Innovation: Mapping the winds of creative destruction. Research Policy 14, 3–22.
Acha, V., Davies, A., Hobday, M., Salter, A., 2004. Exploring the capital goods economy: complex product systems in the UK. Industrial and Corporate Change 13, 505–529.
Bakker, S., van Lente, H., Meeus, M.T.H., 2012. Dominance in the prototyping phase—The case of hydrogen passenger cars. Research Policy 41, 871–883.
Baldwin, C., Maccormack, A., Rusnak, J., 2014. Hidden Structure: Using Network Methods to Map Product Architecture. Research Policy 43, 1381–1397.
Barberá-Tomás, D., Jiménez-Sáez, F., Castelló-Molina, I., 2011. Mapping the importance of the real world: The validity of connectivity analysis of patent citations networks. Research Policy 40, 473–486.
Batagelj, V., Mrvar, A., 2004. Pajek—analysis and visualization of large networks. Springer, Berlin, Germany.
Bekkers, R., Martinelli, A., 2012. Knowledge positions in high-tech markets: Trajectories, standards, strategies and true innovators. Technological Forecasting and Social Change 79, 1192–1216.
Bergek, A., Jacobsson, S., 2003. The emergence of a growth industry: a comparative analysis of the German, Dutch and Swedish wind turbine industries, in: Metcalfe, J.S., Uwe Cantner (Eds), Change, Transformation and Development. Springer, Berlin & Heidelberg, Germany.
BNEF, 2012. Wind Energy Innovation. Bloomberg New Energy Finance (BNEF), New York, NY.
Brusoni, S., Prencipe, A., Pavitt, K., 2001. Knowledge Organizational Coupling, and the Boundaries of the Firm: Why Do Firms Know More Than They Make? Administrative Science Quarterly 46, 597–621.
Campbell, D.T., 1990. Levels of organization, downward causation, and the selection-theory approach to evolutionary epistemology, in: Greenberg, G., Tobach, E. (Eds), Theories of the Evolution of Knowing. Lawrence Erlbaum, Hillsdale, NJ, pp. 1–17.
Castaldi, C., Fontana, R., Nuvolari, A., 2009. “Chariots of fire”: the evolution of tank technology, 1915--1945. Journal of Evolutionary Economics 19, 545–566.
Choi, H., Anadón, L.D., 2014. The role of the complementary sector and its relationship with network formation and government policies in emerging sectors: The case of solar photovoltaics between 2001 and 2009. Technological Forecasting and Social Change 82, 80–94.
88 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Clark, K.B., 1985. The interaction of design hierarchies and market concepts in technological evolution. Research Policy 14, 235–251.
Constant, E., 1973. A model for technological change applied to the turbojet revolution. Technology and Culture 14, 553–572.
Constant, E.W., 1980. The origins of the turbojet revolution. Johns Hopkins University Press, Baltimore, MD.
Davies, A., Hobday, M., 2005. The business of projects: managing innovation in complex products and systems. Cambridge University Press, Cambridge, UK.
Dedehayir, O., Mäkinen, S., 2011. Determining reverse salient types and evolutionary dynamics of technology systems with performance disparities. Technology Analysis & Strategic … 23, 37–41.
Dosi, G., 1982. Technological paradigms and technological trajectories: A suggested interpretation of the determinants and directions of technical change. Research Policy 11, 147–162.
Epicoco, M., 2013. Knowledge patterns and sources of leadership: Mapping the semiconductor miniaturization trajectory. Research Policy 42, 180–195.
Epicoco, M., Oltra, V., Saint Jean, M., 2014. Knowledge dynamics and sources of eco-innovation: Mapping the Green Chemistry community. Technological Forecasting and Social Change 81, 388–402.
Ethiraj, S., 2007. Allocation of inventive effort in complex product systems. Strategic Management Journal 584, 563–584.
Fixson, S.K., Park, J.K., 2008. The power of integrality: Linkages between product architecture, innovation, and industry structure. Research Policy 37, 1296–1316.
Fontana, R., Nuvolari, A., Verspagen, B., 2009. Mapping technological trajectories as patent citation networks. An application to data communication standards. Economics of Innovation and New Technology 18, 311–336.
Frenken, K., 2006. Innovation, Evolution and Complexity Theory. Edward Elgar, Cheltenham UK, Northampton MA.
Frenken, K., Nuvolari, A., 2004. The early development of the steam engine: an evolutionary interpretation using complexity theory. Industrial and Corporate Change 13, 419–450.
Frenken, K., Saviotti, P.P., Trommetter, M., 1999. Variety and niche creation in aircraft, helicopters, motorcycles and microcomputers. Research Policy 28, 469–488.
89
Funk, J., 2009. Components, systems and discontinuities: The case of magnetic recording and playback equipment. Research Policy 38, 1192–1202.
Garrad, A., 2012. The lessons learned from the development of the wind energy industry that might be applied to marine industry renewables. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences 370, 451–71.
Garud, R., Karnøe, P., 2003. Bricolage versus breakthrough: distributed and embedded agency in technology entrepreneurship. Research Policy 32, 277–300.
Gipe, P., 1995. Wind energy comes of age. John Wiley & Sons Inc, New York, NY.
GWEC, 2011. Global Wind Report - Annual Market Update. Global Wind Energy Council (GWEC), Brussels, Belgium.
Hansen, A.D., 2012. Generators and Power Electronics for Wind Turbines, in: Ackermann, T. (Ed), Wind Power in Power Systems. John Wiley & Sons, Inc., New York, NY.
Hau, E., 2013. Wind Turbines - Fundamentals, Technologies, Application, Economics. Springer, Berlin, Germany.
Hayami, Y., Ruttan, V., 1973. Professor Rosenberg and the Direction of Technological Change: A Comment. Economic Development and Cultural Change 21, 352–355.
Henderson, R.M., Clark, K.B., 1990. Architectural Innovation: The Reconfiguration of Existing Product Technologies and the Failure of Established Firms. Administrative Science Quaterly 35, 9–30.
Hendry, C., Harborne, P., 2011. Changing the view of wind power development: More than Briccolage. Research Policy 40, 778–789.
Hidalgo, C., Klinger, B., Barabási, A.-L., Hausmann, R., 2007. The product space conditions the development of nations. Science 317, 482–7.
Hoppmann, J., Peters, M., Schneider, M., Hoffmann, V.H., 2013. The two faces of market support—How deployment policies affect technological exploration and exploitation in the solar photovoltaic industry. Research Policy 42, 989–1003.
Hughes, T.P., 1983. Networks of Power - The Electrification of Western Society, 1880-1930. Johns Hopkins University Press, Baltimore, MD.
Hughes, T.P., 1992. The dynamics of technological change: salients, critical problems, and industrial revolutions, in: Dosi, G., Giannetti, R., Toninelli, P.M. (Eds), Technology and Enterprise in A Historical Perspective. Clarendon Press, Oxford, UK.
90 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Hummon, N.P., Doreian, P., 1989. Connectivity in a citation network: The development of DNA theory. Social Networks 11, 39–63.
Karnøe, P., 1993. Approaches to innovation in modern wind energy technology: technology policies, science, engineers and craft traditions. Center for Economic Policy Research Publication 334.
Lancaster, K.J., 1979. Variety, Equity and Efficiency. Columbia University Press, New York, NY.
Langes, D., 1969. The Unbound Prometheus - Technological Change and Industrial Development in Western Europe from 1750 to the Present. Cambridge University Press, Cambridge, UK.
Lee, J., Berente, N., 2013. The era of incremental change in the technology innovation life cycle: An analysis of the automotive emission control industry. Research Policy 42, 1469–1481.
Martinelli, A., 2012. An emerging paradigm or just another trajectory? Understanding the nature of technological changes using engineering heuristics in the telecommunications switching industry. Research Policy 41, 414–429.
Martinelli, A., Nomaler, Ö., 2014. Measuring knowledge persistence: a genetic approach to patent citation networks. Journal of Evolutionary Economics 24, 623–652.
Menanteau, P., 2000. Learning from Variety and Competition Between Technological Options for Generating Photovoltaic Electricity. Technological Forecasting and Social Change 63, 63–80.
Mendonça, S., 2012. The evolution of new combinations: Drivers of British maritime engineering competitiveness during the nineteenth century. PhD thesis, University of Sussex in Brighton, UK.
Menzel, M.-P., Kammer, J., 2011. Pre-entry Experiences, Technological Designs, and Spatial Restructuring in the Global Wind Turbine Industry. DIME Final Conference 6-8 April 2011, Maastricht, The Netherlands.
Mina, A., Ramlogan, R., Tampubolon, G., Metcalfe, J.S., 2007. Mapping evolutionary trajectories: Applications to the growth and transformation of medical knowledge. Research Policy 36, 789–806.
Mowery, D.C., Rosenberg, N., 1979. The influence of market demand upon innovation: a critical review of some recent empirical studies. Research Policy 8, 102–153.
Murmann, J.P., Frenken, K., 2006. Toward a systematic framework for research on dominant designs, technological innovations, and industrial change. Research Policy 35, 925–952.
Murmann, J.P., Tushman, M.L., 2002. Dominant Designs, Technology Cycles, and Organizational Outcomes, in: Garud, R., Kumaraswamy, A., Langlois, R.N. (Eds), Managing in the Modular Age: Architectures, Networks, and Orgaetworks, and Organizations. Blackwell, Malden, MA.
91
Nemet, G.F., 2009. Demand-pull, technology-push, and government-led incentives for non-incremental technical change. Research Policy 38, 700–709.
Nielsen, K.H., 2010. Technological Trajectories in the Making: Two Case Studies from the Contemporary History of Wind Power. Centaurus 52, 175–205.
Pavitt, K., 1984. Sectoral patterns of technical change: towards a taxonomy and a theory. Research Policy 13, 343–373.
Rosenberg, N., 1963. Technological change in the machine tool industry, 1840–1910. The Journal of Economic History 23, 414–443.
Rosenberg, N., 1969. The direction of technological change: inducement mechanisms and focusing devices. Economic Development and Cultural Change 18, 1–24.
Rosenkopf, L., Nerkar, A., 1999. On the Complexity of Technological Evolution - Exploring Coevolution Within and Across Hierarchical Levels in Optical Disc Technology, in: Baum, J.A.C., McKelvey, B. (Eds), Variations in Organization Science: In Honor of D. T. Campbell. SAGE Publications, Inc.
Sahal, D., 1985. Technological guideposts and innovation avenues. Research Policy 14, 61–82.
Sandén, B.A., Azar, C., 2005. Near-term technology policies for long-term climate targets? Economy-wide versus technology specific approaches. Energy Policy 33, 1557–1576.
Saviotti, P., 1986. Systems theory and technological change. Futures 6, 773–786.
Saviotti, P., Trickett, A., 1992. The evolution of helicopter technology, 1940–1986. Economics of Innovation and New Technology 37–41.
Saviotti, P.P., 1996. Technological Evolution, Variety and the Economy. Edward Elgar, Cheltenham, UK.
Simon, H., 1962. The architecture of complexity. Proceedings of the American Philosophical Society 106, 467– 482.
Teubal, M., 1979. On User Needs and Need Determination Aspects of the Theory of Technological Innovation, in: Baker, M. (Ed), Industrial Innovation. Technology, Policy and Diffusion. Macmillan, London, UK, pp. 266–289.
Tushman, M., Rosenkopf, L., 1992. Organizational determinants of technological-change-toward a sociology of technological evolution. Research in Organizational Behavior 14, 311.
USGS, 2014. Wind Turbine Dataset. US Geological Survey (USGS), available at http://pubs.usgs.gov/ds/817/, Denver, CO.
92 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Verspagen, B., 2007. Mapping technological trajectories as patent citation networks: A study on the history of fuel cell research. Advances in Complex Systems 10, 93–115.
Vincenti, W., 1990. What engineers know and how they know it: Analytical studies from aeronautical history. Johns Hopkins University Press, Baltimore, MD.
Von Wartburg, I., Teichert, T., Rost, K., 2005. Inventive progress measured by multi-stage patent citation analysis. Research Policy 34, 1591–1607.
WWEA, 2012. Small Wind World Report. World Wind Energy Association (WWEA), Bonn, Germany.
93
Appendix
Table A1: Wind-specific engineering tasks and main involved knowledge domains
Components Engineering tasks Main knowledge domains
Rotor
Aerodynamic and structural design of rotor to capture wind energy Aerodynamics; structural dynamics
Design of non-destructive testing equipment and procedures Optics; robotics; mechanical engineering
Development of tailored structural materials and coating Materials science; chemistry Blades Chemical, mechanical and thermal process engineering; automation Processing of large-scale composite components and core materials engineering
Design of equipment and routines for transport and installation of rotor blades Logistics, mechanical engineering
Hub Structural design and integration of O&M and control features Aerodynamics; structural dynamics
Design of rotor control strategy and software Aerodynamics; control engineering, software engineering Rotor control system (pitch and yaw mechanisms), control routines Design and integration of electric motors, gears, hydraulics and power sources Electrical, mechanical, and control engineering
Power train
Design of drive-train architecture Mechanical engineering Mechanical drive-train: Rotor shaft, bearings, gearbox, couplings Dimensioning and material selection for hub, bearings, shafts, brakes, gearbox, lubrication, joints and Material science; structural dynamics couplings
Design of generator topology Electrical engineering, electronics Electrical drive-train: generator, power electronics Design and dimensioning of generator, power electronics, and cooling systems Electrical engineering; electronics, thermodynamics
Design of rotor control strategy and software Aerodynamics; control engineering, software engineering Power-train control system and routines Design and integration of switch board, sensors, actuators (e.g., brakes) and power sources Electrical, mechanical, and control engineering
94 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Table A1 (continued): Wind-specific engineering tasks and main involved knowledge domains
Mounting & encapsulation
Design of load transfer, noise insulation and thermal management Mechanics; acoustics; thermodynamics Nacelle, spinner and bedplate Aesthetic and aerodynamic design Industrial design; aerodynamics
Choice of tower shape, modularity, and structural materials Materials science Tower Dimensioning against bending and fatigue Structural dynamics; mechanical engineering
Foundation Dimensioning for static and dynamic load transfer Mechanics; civil engineering
Thermodynamics; structural dynamics; control and software Design of control strategy and software engineering Climate & vibration control system and routines
Design and integration of dampers, sensors and climate conditioning system Thermodynamics; control engineering
Grid connection
Transformer / substation and power cables Design of wind-farm circuitry, voltage transfer, electrical insulation Electrical engineering
Storage Choice and design of storage technology Electrical engineering, electronics
Design of control strategy and software Electrical, control and software engineering Grid-impact and wind-farm control system and routines Design and integration of control system elements Electronics; control engineering
95
Table A2: Design options within the product architecture of horizontal axis wind turbines operating on the lift principle
Salient design features Design options (today’s most common design in bold)
Wind turbines (system-level) Vertical axis, horizontal axis; drag-based, lift-based energy extraction
Rotor
Rotor position relative to power train and tower Facing the wind (upwind), facing away from the wind (downwind)
Rotor size 5-160 m diameter (~100 m)
Number of blades 1, 2, 3, many
Aerodynamic (‘stall-controlled‘), rotation of blades around own axis to Rotor speed control control lift (‘pitch’), hybrid forms
Rotor orientation control Yaw drive, positioning vane
Glass fiber reinforced plastics, carbon fiber reinforced plastics, wood Rotor material composites, aluminum, steel
Rotor fixation Fixed, hinged, teetered
Power train
Number of bearings 1, 2, 3
Mechanical transmission Gearbox, without gearbox (’direct drive’)
Number of transmission ratios (‘speeds’) 1-5 fixed speeds, variable speed
Number of generators 1-4
5 kW – 7.5 MW (~3 MW) / asynchronous (wound rotor, squirrel cage), Generator size / type synchronous (permanent, wound rotor)
Power converters (rectifier & inverter) Full, partial, none
Mounting & encapsulation
Nacelle / spinner None, reinforced-plastic cover
Tower structure / height Tubular, lattice / 20-130 m (~100m)
Foundation Concrete slab, pile
Grid connection
Storage None, battery storage, compressed-air storage
Grid-integration control None, fault ride-through capability, power control capability
96 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Wind Turbine Technology
Table A3: Goodness-of-fit test of distribution of patent citations from sub-system i to sub-systems j=1…4 with a null hypothesis that the distribution of citations follows the distribution of possible recipient patents (citations to system-level patents were excluded).
Citations from patents categorized into sub-system… N Degrees of freedom χ2 p
Rotor 400 3 271 <0.001
Power train 885 3 318 <0.001
Mounting & encapsulation 651 3 458 <0.001
Grid connection 886 3 555 <0.001
97
Table A4: Design hierarchy, as determined by relationship between components (rows) and un-weighted service characteristics (columns)
Initial cost Reliability & durability Electrical characteristics Environmental impact Others System / sub-systems / Hierarchy of nested parts / Pleiotropyb components Turbine Cost pf transport, Availability& Power Grid Visual Noise Operational Suitable climate Hierarchy of controlc Lifetime cost installation & disassembly O&M cost curve behavior impact emissions safety conditions
Wind turbine (system-level) Xa X X X X X X X X X 10 1
Rotor XX XXXXX XXX102/A
Rotor blades X X X X X X X X X X 10 3
Hub X XX X X 5 3
Rotor control X XXXX XXX83
Power train XX XXXX XXX92/B
Mechanical drive-train X X X X X X X X X 9 3
Electrical drive-train X X X X X X X X 8 3
Power-train control XXXX XX X 7 3
Mounting & encapsulation XX XX X XXX82/C
Nacelle, spinner & bedplate X X X X X X 6 3
Tower XX X X 4 3
Foundation X X 2 3
Climate and vibration X X X X 4 3 control
Grid connection X X XX XX 62/D
Transformer / substation X X X X X X 6 3 and power cables
Storage (if applicable) X X X X 4 3
Wind-farm and grid- X X X X 4 3 integration control aEach x represents a significant influence of the respective sub-assembly or the individual component (in rows) on the main service characteristic (in columns); bThe pleiotropy is the count of influences per row. cThe number of the design hierarchy indicates the hierarchy of nested parts (1=system, 2=assembly of components, 3=component); the capitalized letter indicates the hierarchy of control on each level (A=highest pleiotropy, B…D sorted accordingly)
98 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge–Evidence from Wind Turbine Technology
Table A5: Patents along critical path of wind-patent citation network 1973-2009
Priority Focus in Application Focus of invention Assignee Assignee type patent hierarchy
SE Svenning Engineering 12-May-75 Blade with integrated over-speeding control mechanism Rotor 005,407 Konsult AB consultancy
DE U. Huetter Public sector 4-Dec-76 Rotor-hub arrangement with teetering hub and two blades Rotor 2,655,026 (Indiv.) (university)
US Turbine 8-Jun-78 Control system for two-bladed rotor with adjustable tips Rotor MAN 4,297,076 manufacturer
US C E Kenney 31-Jul-78 Three-bladed turbine with hydraulic pitch mechanism Rotor Individual 4,274,807 (Indiv.)
US Two-bladed downwind turbine with teetering hub and aerodynamic Carter Wind Turbine 10-May-79 Rotor 4,366,387 pitch mechanism Power manufacturer
US North Wind Turbine 24-Feb-82 Rotor with teetered hub and mechanical pitch control system Rotor 4,435,646 Power manufacturer
US Turbine 29-Sep-83 Two-blade turbine with novel drag brake and control system Rotor Boeing 4,565,929 manufacturer
US United Turbine 18-Nov-85 Torque control system for variable-speed power train Power train 4,703,189 Technologies manufacturer
US United Turbine 28-Apr-86 Operation strategy for variable-speed power train Power train 4,700,081 Technologies manufacturer
US US Turbine 1-Feb-91 Variable-speed power train architecture and power control Power train 5,083,039 WindPower manufacturer
US US Turbine 19-Sep-91 Speed control system for variable-speed power train Power train 5,155,375 WindPower manufacturer
Public sector US 6-Feb-95 Fuzzy-logic power train control for variable wind conditions Power train U.S. EPA (regulatory 5,652,485 agency)
US Zond Energy Turbine 8-Aug-97 Variable-speed power train architecture and power control Power train 6,137,187 Systems manufacturer
US Vestas Wind Turbine 23-May-00 Variable-speed power train adapted to smoothen power output Power train 6,566,764 Systems manufacturer
Component US Grid 10-Jul-01 Inverter control system for grid-friendly power output ABB supplier 6,670,721 connection (generator)
DE Grid Turbine 28-Sep-01 Collective control method for turbines in a wind farm Enercon 1,048,225 connection manufacturer
Component US 8-Apr-03 Variable-speed power train architecture Power train Alstom supplier 7,190,085 (generator)
US Clipper Turbine 7-May-03 Variable-speed power train architecture Power train 7,042,110 Windpower manufacturer
US Grid Turbine 8-Jan-04 Generator control optimizing response to grid failure Hitachi 7,205,676 connection manufacturer
Mounting & Mitsubishi Turbine JP 055,515 27-Feb-04 System to control nacelle vibrations encapsulation Heavy Ind. manufacturer
99
US Mounting & General Turbine 30-Sep-04 System to control turbine vibrations 7,309,930 encapsulation Electric manufacturer
US General Turbine 30-Sep-05 Power train control routine based on upstream wind measurements Power train 7,342,323 Electric manufacturer
US Mounting & Fuji Heavy Turbine 1-Feb-06 Control routine to suppress tower vibrations 7,400,055 encapsulation Industries manufacturer
US Grid Turbine 14-Sep-06 Control routine to respond to grid faults Vestas 7,851,934 connection manufacturer
US Grid Turbine 14-Sep-06 Control routine to respond to grid faults Vestas 7,911,072 connection manufacturer
US Grid Turbine 22-Feb-08 Control routine to respond to grid‐side load shedding Nordex 7,714,458 connection manufacturer
US Grid Turbine 16-Jun-08 Control system for wind farm with redundant control unit Nordex 7,949,434 connection manufacturer
100 How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge–Evidence from Wind Turbine Technology
Table A6: Shifting focus in hierarchy of control along trajectory of knowledge generation, indicated by share of vertex weight in 95%-weight network in different elements of the system (number of patents in parentheses)
1975-1979 1980-1984 1985-1989 1990-1994 1995-1999 2000-2004 2005-2009
Wind turbine 0.02 (2) 0.00 (1) 0.04 (3) 0.07 (3) 0.08 (3) 0.00 (2) 0.01 (4) (system-level)
Rotor 0.77 (12) 0.76 (17) 0.05 (7) 0.04 (2) 0.01 (3) 0.02 (11) 0.18 (60)
Power train 0.13 (5) 0.16 (5) 0.91 (8) 0.87 (8) 0.78 (14) 0.34 (39) 0.23 (67)
Mounting & 0.04 (1) 0.05 (3) 0 (0) 0.02 (2) 0.03 (8) 0.18 (20) 0.32 (67) encapsulation
Grid connection 0.04 (1) 0.04 (1) 0 (0) 0 (0) 0.10 (6) 0.46 (4) 0.25 (75)
101
Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation‡
Joern Huenteler1,2*, Tobias S. Schmidt1,3, Jan Ossenbrink1, Volker H. Hoffmann1
1Department of Management, Technology and Economics, ETH Zurich, Switzerland 2Belfer Center for Science and International Affairs, John F. Kennedy School of Government, Harvard University, USA 3Precourt Energy Efficiency Center, Stanford University, USA *corresponding author: [email protected]; +41 44 632 97 39; +41 44 632 05 41
‡Working paper, as of December 7, 2014. Previous versions of this paper have been presented at the Energy Policy Consortium Seminar at Harvard in 2014, the UNFCCC COP 18 in Doha, Qatar, the International Sustainability Transitions 2012 conference in Copenhagen, Denmark and the International Schumpeter Society Conference 2012 in Brisbane, Australia.
Abstract
Understanding the long-term patterns of innovation in energy technologies is crucial to inform public policy planning in the context of climate change. We analyze which of two common models of innovation over the technology life-cycle – the product-process innovation shift observed for mass-produced goods or the architecture-component shift observed for complex products and systems – best describes the pattern of innovation in energy technologies. To this end, we develop a novel, patent-based methodology to study how the focus of innovation changed over the course of the technology life-cycle. Specifically, we analyze patent-citation networks in solar PV and wind power in the period 1963-2009. The results suggest that solar PV technology followed the life-cycle pattern of mass-produced goods – early product innovations were followed by a surge of process innovations in solar cell production. Wind turbine technology, in contrast, more closely resembled the life-cycle of complex products: the focus of innovative activity shifted over time through different parts of the product, rather than from product to process innovations. These findings indicate very different innovation and learning processes in the two technologies and the need to tailor technology policy to technological characteristics, and help conceptualizing previously inconclusive evidence about the impact of technology policies in the past.
Keywords: Technology life-cycle, energy technology, patents, citation network analysis, wind power, solar PV
104 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
Highlights:
We develop a patent-based methodology to study technology life-cycles. We apply the methodology to solar PV and wind power over the period 1963-2009. PV followed the life-cycle of mass-produced goods and commodities. Wind power followed the life-cycle of complex products and systems. We develop a typology of energy technologies and discuss policy implications.
105
1. Introduction
Technological change is “at once the most important and least understood feature driving the future cost of climate change mitigation” (Pizer and Popp, 2008, p. 2768). Better understanding the long- term patterns of innovation in energy technologies is therefore crucial to inform public policy planning (Grubb, 2004; Pielke et al., 2008; Grubler, 2012), and a growing body of literature is studying innovation processes and technology policy in the energy sector (e.g., Anadon, 2012; Gallagher et al., 2012; Grubler and Wilson, 2014). It is a particularity of the energy sector that technologies from a diverse range of sectors of the economy are employed in the extraction, conversion and end-use of energy. Therefore, most energy innovations are not developed by energy utilities, but enter the sector embodied in specialized equipment or innovative fuels from other sectors, such as semiconductors (solar cells), electro-mechanical machinery (gas turbines), agriculture (biofuel feedstock) and biochemistry (biofuel conversion) (Markard, 2011; Wiesenthal et al., 2011). Some of these energy technologies are large capital goods, such as biomass power plants, while others are modular and produced in large volumes, such as heat pumps (Neij, 1997). Some are very specific to local geographies, such as geothermal power plants, while others are globally applicable almost without adaptation, such as solar cells (Huenteler et al., 2014a). Yet so far only few studies have investigated systematically how innovation processes differ between energy technologies originating in different sectors, and explored the implications for energy technology policy (Norberg-Bohm, 2000; Trancik, 2006; Wilson, 2012; Winskel et al., 2014).
There is a rich empirical literature describing how innovation processes differ between sectors in general (Pavitt, 1984; Marsili, 2001), with a particular focus on different temporal patterns of innovation (Rosenberg, 1982; Dosi, 1988; Dosi and Nelson, 2013). For example, cross-sectoral comparisons present evidence that patterns of innovation over the technology life-cycle differ between mass-produced goods and complex capital goods (Hobday, 1998; Davies and Hobday, 2005; Magnusson et al., 2005). In mass-produced goods, the focus of innovative activity tends to shift over time from product to process innovations as one dominant design is adopted throughout the industry and the general thrust of R&D shifts toward increasing automatization of the production process (Vernon, 1966; Abernathy and Utterback, 1988; Peltoniemi, 2011). This is accompanied by an increasing role for learning-by-doing in production over time (Hatch and Mowery, 1998). In complex products and systems, such as aircraft, trains or nuclear plants, in contrast, a shift toward large-scale production is extremely rare; instead, the focus of innovative activity tends to shift from the system architecture
106 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation towards individual sub-systems or components (Miller et al., 1995; Davies, 1997; Murmann and Frenken, 2006). In these cases, learning-by-using and user-producer interaction remain important even for mature technologies (Rosenberg, 1982).
The two contrasting models of innovation in mass-produced goods and complex products and systems describe, in a stylized way, different drivers of innovation and mechanisms of learning in later stages of the life-cycle. They thus have important implications for the design of technology policies for clean energy technologies, many of which seek to induce innovations and cost reductions in relatively mature technologies – most clean technologies supported today, including wind, solar and biomass, have been first deployed at scale in the wake of the oil crises in the 1970s and 80s and are now in later stages of the technology life-cycle. However, much of the technology policy debate today on the on innovation in these technologies is centered on learning-by-doing in manufacturing and economies of scale, reflecting a mental model heavily influenced by the life-cycle model for mass-produced goods.28
Most of the currently supported clean energy technologies are somewhere between the two extremes. Unlike commodities and mass-produced consumer goods, the designs of wind turbines, solar PV systems, fuel cells or electric vehicles all feature a number of different sub-systems and components, and continue to evolve even after their first large-scale deployment. But unlike complex products and systems they are also produced in significant volumes. In order to stimulate innovation in these technologies effectively, we need to better understand the processes of innovation in later stages of the technology life-cycle. However, researchers have so far paid relatively little attention to the question whether patterns of innovation differ significantly over the life-cycle of technologies in the energy sector. To address this gap, this paper develops a patent-based methodology to analyze which of the two models of the technology life-cycle – the product-process innovation shift observed for mass- produced goods or the architecture-component shift observed for complex products and systems – best describes a pattern of innovation over time and apply this methodology to the two most rapidly growing clean energy technologies: solar photovoltaics (PV) and wind power.
28 For example, the German feed-in tariff for solar power (a form of subsidized electricity tariff), with about $10bn per year currently the largest deployment policy in the world, was designed as "market entry assistance to allow for cost reductions, which will then facilitate the diffusion of photovoltaic through the market" (German Federal Diet, 1999). The US tax credit under the U.S. ‘Recovery Act’ in 2009 had the objective “to help renewable energy technologies achieve economies of scale and bring down costs” (The White House, 2009).
107
The paper proceeds as follows. Section 2 introduces two alternative models of the technology life-cycle and discusses the main technological determinants of life-cycle patterns discussed in the literature. Section 3 introduces the two case technologies – solar PV systems and wind turbines – and presents key indicators of progress along the technological trajectory, such as cost and efficiency, over the last five decades. In section 4, we introduce a novel methodology to study how the focus of innovative activity evolved along the trajectories of the two case technologies. The results are presented in section 5 and their implications discussed in section 6. Section 7 summarizes the main conclusions.
2. Theoretical Perspective and Literature Review
The ‘life-cycle’ metaphor has been used in many different contexts in research on the management and economics of innovation (Routley et al., 2013). This paper draws on the literature that uses the term life-cycle to describe the temporal patterns of technological innovation in an industry, in particular the emergence of dominant designs and architectures and the corresponding shifts in the focus of innovation, observed across a wide range of technologies (Utterback and Abernathy, 1975; Abernathy and Utterback, 1978; Clark, 1985; Utterback and Suárez, 1993; Murmann and Tushman, 2002; Murmann and Frenken, 2006; Lee and Berente, 2013).
2.1. Two Contrasting Models of the Technology Life-Cycle
Technological evolution in manufactured products often takes a cyclical form, with an early stage marked by intense product innovation and competition among fundamentally different product concepts (Anderson and Tushman, 1990; Suarez and Utterback, 1993; Murmann and Frenken, 2006). After a dominant design has emerged, technological change becomes cumulative and incremental as innovation proceeds along ordered technological trajectories (Dosi, 1982; Mina et al., 2007; Verspagen, 2007; Fontana et al., 2009; Bekkers and Martinelli, 2012).
The most influential model of temporal patterns of innovation holds that technological trajectories are punctuated by technological discontinuities which initiate cycles of product and process innovation (e.g., Vernon, 1966; Utterback and Abernathy, 1975; Abernathy and Utterback, 1978, 1988; Utterback and Suárez, 1993). Initially, the focus of the innovative activity in the industry is on product innovation, as firms try to exploit the performance potential of the discontinuous innovation and compete in the market with many alternative product designs. This ‘era of ferment’ culminates in a dominant design as the technology’s core components become standardized. What follows is an ‘era of incremental
108 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation change’, during which the technology proceeds along defined trajectories and the focus of innovative activity is on process innovations and specialized materials, as firms compete primarily on the basis of costs – until a new discontinuity re-ignites design competition (see Figure 1a). The shift from product to process innovations is due in part to standardization of product design feature, but also the result of a transition from the era of ferment to the era of incremental change is also characterized by a shift from small-batch production to mass production, and from general-purpose plants to large manufacturing facilities with highly specialized production equipment (see Table 1) (Abernathy and Utterback, 1988).
The Abernathy-Utterback (A-U) model has been extremely influential29, but several authors, including Miller et al. (1995) and Davies (1997), note that the model is valid only for a subset of technologies. In particular, empirical studies of innovation in complex, capital-intensive goods demonstrate that many high-value, high-technology products, so-called complex products and systems, never reach a phase of process innovation and mass production and innovative activity remains focused on product innovation throughout the life-cycle (see Table 1) (Davies, 1997; Hobday, 1998; Davies and Hobday, 2005). This is in line with studies of the era of incremental change which found in many cases little evidence for a decline in product innovations (Gort and Klepper, 1982; Henderson, 1995; Lee and Berente, 2013).
Based on this evidence, Davies (1997) introduces a model of innovation over time that characterizes the life-cycle of complex products as one where the product-process shift observed for mass-produced goods is replaced by a shift from innovation in the system architecture to waves of innovation in sub- systems and components (see Figure 1b) (Davies, 1997; Davies and Hobday, 2005). As in the A-U model, the early phase is characterized by a focus on functional performance and product innovations. However, the competitive emphasis is not on specific designs but on innovations in the product architecture, which allocate system functions to the individual components and defines the interfaces between them (e.g., Clark, 1985). After the emergence of a dominant product architecture and standardized core sub-systems (the dominant design), innovation along the technological trajectory is focused on individual sub-systems and components (Murmann and Frenken, 2006). For example, after the emergence of the turbojet engine as the dominant propulsion system, innovative activity in the aircraft industry focused on improving the airframe and parts of the engine, such as compressor blades,
29 The two seminal works (Utterback and Abernathy, 1975; Abernathy and Utterback, 1978) had a total of 6,544 google scholar citations between them on 12/6/2014.
109
rather than shifting toward process mechanization and automatization (Constant, 1980). Over time, changes in sub-systems and components may create performance imbalances and thus require changes in other parts of the system (Brusoni et al., 2001; Funk, 2009), in which case Davies refers to them as ‘systemic innovations’ (see Figure 1b).
a) Mass-produced products and commodities b) Complex products and systems
Rate of major Rate of major innovation innovation Component/systemic innovation Product innovation Architectural innovation Process/ innovation
Time Time Dominant design Dominant architecture emerges emerges
Era of ferment Era of incremental change Era of ferment Era of incremental change
Figure 1: Two contrasting models of innovation over the technology life-cycle: a) mass-produced goods; b) complex products and systems (Abernathy and Utterback, 1988; Davies, 1997).
Era of incremental change Era of ferment Mass-produced goods Complex products and systems
Competitive Functional product performance Cost reduction Functional product performance emphasis on …
Innovation Revealed user needs and Pressure to reduce cost and improve Evolving user needs as well as internal and external stimulated by … users‘ technical inputs quality technical opportunities
Diverse, often including custom Mostly undifferentiated standard Product variations that share common architecture but Product line designs products are customized to user needs
Predominant type Frequent major product Incremental innovation in processes and Sequences of systemic and incremental component of innovation innovations materials changes
Important sources Product R&D, learning-by- Process R&D, learning-by-doing Product R&D, learning-by-using of knowledge doing and learning-by-using
Large-scale plant tailored to particular General-purpose plant with specialized sections General-purpose plant located Plant product designs to realize economies of located near user or source of technology, little near user or source of technology scale emphasis on economies of scale
Flexible and inefficient: major Efficient, capital-intensive. and rigid: Remains flexible: individual projects or small-batch Production process changes easily accommodated cost of change is high production
Special-purpose, mostly automatic with Production General-purpose equipment, Some sub-processes automated, but mostly requiring labor tasks mainly monitoring and equipment requiring highly skilled labor highly skilled labor control
Table 1: Characteristics of the innovation and production processes in the two alternative models of the technology life-cycle (Abernathy and Utterback, 1988; Davies, 1997).
110 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
The two models differ most significantly in their characterization of innovation along the technological trajectory after a dominant design or product architecture has emerged (see Table 1). Three aspects are particularly important to characterize innovation and learning processes in the era of incremental change: First, with regard to the type and breadth of innovative activity, the A-U model predicts a surge in process innovations and a relatively narrow focus on cost reductions through improved production processes. The Davies model describes a steady stream of product innovations as well as a broadening of the focus from the system architecture and core sub-systems to a broader range of sub-systems and components, with an emphasis on understanding and enhancing the complex interactions between different elements of the system. Second, the A-U model ascribes an important role to the exploitation of economies of scale to realize cost reductions, which implies a role for learning- by-doing in manufacturing (e.g., Hatch and Mowery, 1998). Davies’ model, in contrast, sees the later stage of the life-cycle as still characterized by small-scale, flexible production plants that allow limited learning-by-doing and economies of scale. And third, with regard to the role of performance uncertainty and learning-by-using, the A-U model predicts a rapid decline in uncertainty about the functional performance of different design features and user needs. This results in very little need in the innovation process for experience from large-scale or long-term experimentation and user-producer interaction, which allows moving factories to locations with cost advantages even if they are far from the actual users (e.g., Vernon, 1966). This is in stark contrast to the continued dependence on learning-by-using and close proximity between users and producers that characterizes complex products and systems (e.g., Rosenberg, 1982).
2.2. Technological Characteristics and Life-Cycle Patterns
How can one locate specific technologies in the continuum created by the two described life-cycle models? In the literature on complex products and systems, the term ‘complex’ refers to a large number of drivers, e.g., the number of customized components, the scale of the product and the intensity of regulatory involvement in the specification of requirements (Hobday, 1998). When comparing the patterns of innovation over the life-cycles, Davies (1997) narrows these drivers down to four main characteristics of complex products and systems vis-à-vis mass-produced goods: (i) the complexity of product architecture, (ii) the scale of the production process, (iii) the market structure (oligopoly versus mass market) and (iv) the degree of government involvement in technological evolution (which is often unusually high for complex products and systems).
111
When applying these characteristics to the energy sector, these determinants can be further reduced to two underlying technological characteristics. On the one hand, the evolution of all energy technologies are heavily affected by government policies, e.g., in the form of technology standards, environmental regulations, subsidy schemes and industrial policy. On the other hand, for energy technologies the scale of the production process is highly correlated with the market structure, since low-volume technologies are typically procured by large, regulated utilities (wind power plants, electricity grids), indicating a bilateral oligopoly, whereas mass-produced energy technologies are mostly used by consumers, either in the form of end-use technologies (e.g., heating systems or electric cars) or as decentralized, small scale energy systems (solar PV systems, solar water heaters). This leaves two main technological determinants of life-cycle patterns in the energy sector:
1. The complexity of the product architecture, understood here as a driven by the number of sub- systems and components and the complexity of their interactions in the system. On the one hand, a complex product architecture implies many opportunities to improve individual elements and their interaction after the emergence of a dominant design. At the same time, architectural complexity is a driver of iterations and learning-by-using in the innovation process, because it makes performance features of the final product difficult to predict (Rosenberg, 1982; Nightingale, 2000). 2. The scale of production process, which is mainly driven by the modularity of the system as well as the size and homogeneity of user demand. A large process scale implies many opportunities to improve cost and functional performance through process innovations. At the same time, it often requires a prolonged process of experimentation and learning-by-doing to develop and operate the large-scale production systems with many interdependent process steps (e.g., Hatch and Mowery, 1998).
The two characteristics span a technology space in the energy sector, with the two life-cycle models as two extremes (see Figure 2). However, the models have been developed based on contrasts between vastly different technologies (e.g., infrastructure systems versus light bulbs), while most energy technologies have relatively complex designs and are produced in non-trivial numbers – i.e., fall somewhere in between the extremes. It is therefore not entirely clear where different types of energy technologies are located on the displayed continuum. In the following sections this paper analyzes two technologies with the aim to locate them in the matrix displayed in Figure 2. We show that important characteristics known for the A-U model and the Davies model can be observed when dissecting the innovation patterns in energy technologies over time.
112 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
High Davies model of life-cycles in complex products and systems Complexity of product architecture A-U model of life-cycles in mass- produced products and commodities
Low
Low High Scale of production process
Figure 2: Technology space in the energy sector, spanned by the scale of the production process and the complexity of the product architecture.
3. Research Cases
This paper explores if life-cycles patterns differ significantly between technologies in the energy sector. The cases analyzed for this purpose need to fulfil two main criteria. First, they need to differ in the two determinants of life-cycle patterns identified above: the complexity of the product architecture and the scale of the production process. Second, they need to be in the era of incremental change, during which the differences we seek to identify become salient.
Wind power and solar PV were selected because they fulfil these criteria: They exhibit different degrees of complexity and different scale of production, as will be discussed in section 3.1. And both have a dominant design and are now in the era of incremental change (see in section 3.2). Last but not least, the two cases are highly relevant for public policy: solar PV and wind power are projected to receive $1.7 trillion and $1.1 trillion in subsidies, respectively, over the period 2013-2040 (IEA, 2014). A better understanding the processes of innovation and technological evolution in these technologies can therefore inform important technology policy decisions in the coming decades.
3.1. Characteristics of the Case Technologies
To delimit the empirical scope of our analysis, we understand the term ‘technology’ as describing a class of artifacts defined by a common ‘operational principle’ and the pertaining procedures and elements of knowledge (Polanyi, 1962; Vincenti, 1990; Murmann and Frenken, 2006). We considered solar PV to include all technology related to power generation using the photovoltaic effect, and wind power to include all technology using lift forces of the wind to generate electricity. Table 2, which
113
presents the elements of solar PV and wind power systems and their functions in the system, illustrates how the central role of these two physical principles in the functional structure of the electricity generation systems. A dominant design is understood here as a standard in design of the technology’s core components (Murmann and Frenken, 2006), which we define here as the rotor in a wind turbine and the cell concept of a PV system.
When comparing the technology characteristics that indicate patterns in the technology life-cycle (see section 2.2) between the two technologies, it becomes evident that the complexity of the product architecture is significantly higher for wind turbines, while the scale of the production process is higher in the case of solar PV.
Solar PV systems are modular systems that consist of small generating units – the solar cells – which are connected to modules of around 200W and integrated with mounting and tracking structures and, in order to feed the electricity into the grid, inverters and control systems (see Table 2). Solar modules have only few components and no moving parts, and currently cost about 150-250$ at the factory gate, depending on the exact capacity rating, efficiency, and other features such as warrantees. The fact that a solar module contains few moving parts is reflected in a very low value of operation and maintenance (O&M) cost, which are often below 1% and rarely exceed 5% (Moore and Post, 2008). Solar cells are produced, in batches of at least several thousands, on large and specialized, automated production lines which cost up to several billion USD. Consequently, the market for solar modules exhibits many features of mass-manufactured commodities – even spot markets for cells and modules (e.g., Barua et al., 2012).
Modern wind turbines, in contrast, are electro-mechanical machines that can reach up to 8 MW of electric capacity, consist of several thousand components and cost up to $15 million per unit (a list of key sub-systems and main functions is given in Table 2). Although they are typically not made-to- order, wind turbines often contain site specific characteristics, such as sand and dust in the air, high altitude sites or very cold climate. The high number of moving key components is reflected in high O&M cost, which make up 20-25 % of the cost of electricity over the lifetime of a wind turbine (Twidell, 2009). Wind turbine production and construction processes are dominated by what one of our interviewees called “simple industrial craftsmanship”, i.e., standard industrial processes that require skilled manual labor and are performed on multi-purpose machinery, such as welding, milling and drilling machines. Specialized equipment is used only in the blade manufacturing and installation processes, in the form of large moulds and cranes. Overall, a wind turbine production facility has
114 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation construction cost in the order of $20-200 million, depending on annual production capacity, and can produce up to several 100 MW of turbines per year. Further detail is given on both technologies in Table A1 and Table A2 in the appendix, which show the main engineering tasks in the two technologies, as well as the main areas where a technology-specific body of knowledge has emerged.
Table 2: Product architectures of solar PV and wind power systems, showing the main sub-systems and their function in the technological system.
System System element Function
Solar cell Absorption of solar irradiation and conversion into electric current through photovoltaic effect
Connection of ‘string’ of cells to achieve desired output voltage; protection of cell from Solar module moisture and structural damage; insulation of electrical current
Solar PV system Integration of modules into larger structures (array); load carrying and transfer (mounting Mounting system system); integration of module / cells into building environment (building integration); reorientation of modules / array to follow the sun (tracking system)
Conversion of DC current into AC (inverter); reduction of impact of grid-side disturbances; Grid connection maintenance of grid-friendly system output (electrical control system)
Conversion of wind energy into rotational energy through lift effect (rotor blades); transfer of Rotor energy to main shaft (hub); adjustment of rotor and individual blades to wind & system conditions (rotor control system)
Transmission of rotational energy from rotor to generator, including adjustment of rotational frequency (mechanical drive train); conversion of rotational energy into electrical energy, AC- Power train DC conversion and frequency conversion (electrical drive train); adjustment of power-train elements to wind & system conditions (power-train control) Wind power system Load carrying and machinery enclosure (nacelle, spinner, bedplate); support turbine at designated height and load transfer to foundation (tower); load transfer into ground Mounting & encapsulation (foundation); regulation of operating conditions &minimization of system vibrations (climate and vibration control)
Transfer of electrical energy to grid (transformer/substation, power cables); storage of Grid connection electrical energy (storage system, if applicable); reduction of impact of grid-side disturbances; maintenance of grid-friendly wind farm output (grid-impact and wind-farm control)
3.2. Dominant Designs and Technological Trajectories in Solar PV and Wind
Power
This section presents evidence for the fact that both technologies gone through different stages of the life-cycle and are now the era of incremental change, by demonstrating (i) the presence of dominant designs in solar PV and wind power and (ii) the maturity of the industries and the prevalence of cumulative and incremental innovation.
115
The market for solar PV and wind power systems has grown exponentially over the last four three decades. As a comparable indicator for market growth, Figure 3b shows annual installations for the two technologies (very roughly, one MW of installations translates into $1-10m of investment). The data illustrates that both industries have now grown to large and mature industries: In 2012, the PV industry recorded sales of around $80bn and the wind industry of around $75bn (Pernick et al., 2013). Wind power had a head start with early installations in California during the 1980s, and a continued boom since around 1995. But the market for solar PV has caught up in the last decade, and now exhibits similar absolute annual installations rates. The IEA reckons total spending on deployment subsidies to be in the range of $25bn for solar PV and $21bn for wind power (IEA, 2013). The maturity of the industries is further demonstrated by the high relative share of corporate R&D expenditures of total R&D in the two industries, which stands at 58% in solar PV and 76 % in wind power (Wiesenthal et al., 2011).
With the growing market, dominant designs emerged in both industries in the early 1990s (solar PV) and late 1980s (wind power), as shown in Figure 3b. For solar PV, the chart displays market shares by shipment volume (in MW), showing that designs based on wafers of silicon have dominated the market (mono-Si, multi-Si, and ribbon-Si, collectively referred to as crystalline silicon) since the beginning of the industry. Sales of thin-film modules rose during the 1980s when the first commercial-scale installations were financed, and again slightly in the late 2000s, because firms believed that the lower material cost facilitated through the thinner semiconductor would allow thin- film cells to become cheaper than crystalline silicon in the long run. However, both trends were relatively quickly reversed, so that since 1993 the share of crystalline silicon cells has never fallen below 80% of the global market share. Most commercial firms produce the dominant crystalline silicon design, and the technology can thus be considered in the era of incremental change.
For wind power the figure shows trends in the number of companies active in the market pursuing different design concepts. The graph illustrates that the ‘Danish Design’, featuring three-bladed, wind-facing, horizontally mounted rotors, has come to dominate the industry since the late 1980s, when the end of policies in California resulted in a massive shake-out of firms producing light-weight turbines (Menzel and Kammer, 2011). The era of incremental change began with the emergence of the dominant design around 1987, when more than 60% of firms produced the ‘Danish design’ and Danish firms held around 80% of the global market share (Garud and Karnøe, 2003). The Danish design is characterized by a rotor that (a) faces toward the incoming wind, (b) features three rotor blades and (c) operates with relatively low rotational speeds. The dominance of the Danish design has
116 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation only increased since then, albeit with different designs of the transmission system (notably variable- speed gearboxes and gearless transmissions).
a) Market growth Solar PV Wind power 100,000 100,000 10,000 10,000 1,000 1,000 100 100 10 10 1 1 0,1 0,1 1980 1985 1990 1995 2000 2005 2010 19801985 1990 1995 2000 2005 2009
Global annual capacity additions (1 MW ~ 1-10 million $ investment) b) Design competition
Solar PV (market share) Wind power (share of firms)
100% 100% 80% 80%
60% 60% 40% 40%
20% 20% 0% 0% 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2009
Thin film Other Crystalline silicon (Ribbon-Si) Light-weight Crystalline silicon (Multi-Si) Danish design, fixed speed gearbox Crystalline silicon (Mono-Si) Danish design, variable-speed gearbox Danish design, gearless transmission
Figure 3: a) Annual installations of wind power systems (Peters, 2011, p. 132) and solar PV systems (IEA, 2012); b) Design competition in solar PV, as measured by market share of different designs (Fraunhofer ISE, 2012), and in wind power, as measured by the share of firms with different designs active in the market (Menzel and Kammer, 2011).30
Technological change within the dominant designs has been cumulative and incremental over the last three decades, indicating an era of incremental change. Two prominent indicators of technological change in electricity technologies are investment cost31 for new installations (which reflect equipment prices) and efficiency. Both trends are shown in Figure 4a for crystalline silicon PV modules and Danish-
30 The design data for the wind industry does not track design changes, i.e., firms are marked in the database with the design they entered the industry with. The displayed evolution therefore underestimates the rise of variable-speed turbine models, which was adopted by many firms who began with the Danish design. (Firms rarely switched between the other designs.) 31 Since fuel costs do not apply and operation and maintenance are comparatively low, investment costs dominate the economics of renewable electricity.
117
design wind turbines. Figure 4a shows ‘experience curves’, i.e., logarithmic unit prices over the logarithmic cumulative production. The data illustrate that initial prices came down gradually over the last decades. Cumulatively, the effects are very significant: For PV modules, prices in real terms decreased from 76$ per Watt of electric capacity ($/W) in 1976 to around 0.5-1 $/W in 2012. The recent plateau and drop in prices is no indicator of a technological discontinuity; it was mainly driven by imbalances between supply and demand in the markets for raw silicon and solar modules (Candelise et al., 2013). The investment costs of wind farms decreased from around 3.6 $/W to less than 2 $/W in 2011 (BNEF, 2012a, 2012b). Here, too, deviations from the long-term price trend are driven more by factors other than technological change (Bolinger and Wiser, 2012).
a) Investment costs
PV modules Onshore wind turbines Prices [$2011/W] Prices [$2011/W] 100,0 100,0 1976
10,0 Crystalline silicon 10,0 Germany and Denmark
1984 2000 2003
2004 2011 1,0 2011 1,0 Global
0,1 0,1 1 100 10.000 1.000.000 1 100 10.000 1.000.000 Cumulative production Cumulative production b) Conversion efficiency
Average commercial module efficiencya Average turbine capacity factorb Conversion efficiency [%] Capacity factor [%] 20 40 14.7% 34% 15 30
10 20 Data ISE, 2012 21% 5 Data from Nemet, 2006 10
0 0 1980 1985 1990 1995 2000 2005 2010 2015 1980 1985 1990 1995 2000 2005 2010 2015
Figure 4: Technological change within the dominant designs in wind power and solar PV: a) Investment cost (BNEF, 2012a, 2012b); b) PV module conversion efficiency (Nemet, 2006; Fraunhofer ISE, 2012) and wind turbine capacity factors (BNEF, 2012b). Unlike conversion efficiency, the capacity factor includes improvements in siting. The break in the trends for cell efficiency is due to different data sources.
At the same time, suppliers were able to gradually increase the technology quality of the power generation equipment. Figure 4b compiles data for quality indicators commonly used by industry. For solar PV, it shows the average efficiency of commercial PV modules, which increased by a factor of around 1.7 since 1980. The increasing module efficiency reflects incremental reduction of losses in
118 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation many parts of the module and the cell, e.g., through improvements in cell materials, cell treatment, contact printing, contact materials, antireflective coatings, cell interconnection, etc. A similar increase – by a factor of 1.6 - can be seen in the average turbine capacity factors of newly built wind turbines, which relates the actual electricity generation to the maximum possible electricity generation per year. Capacity factors, too, reflect a range of incremental improvements, including the siting of turbines, larger turbine rotors, higher towers, variable-speed transmission systems, and improvements in control systems, thus yielding a comprehensive picture of qualitative progress in wind power. Patent applications grew exponentially in both technologies since the early 1990s, and now stand at several thousand per year (see Figure 5 below). This surge in patenting is consistent with typical patterns in the era of incremental change (Gort and Klepper, 1982; Lee and Berente, 2013).
4. Data and Methodology
4.1. Empirical Strategy
Section 3 provided evidence for the fact that both solar PV and wind power went through different stages of the technology life-cycle. However, the presented indicators offer little cues about the focus of innovative activity, and they leave unanswered whether the patterns conform to one or another model of the technology life-cycle.
This section introduces our patent-based methodology to study the technology life-cycles in wind power and solar PV. Patents have been used extensively to study trends in innovation in technological systems (e.g., Fleming and Sorenson, 2001; Rosenkopf and Nerkar, 2001), in part because they are readily available as large empirical datasets. However, large patent datasets make in-depth analyses difficult – such as the identification of product and process patents – while only containing a small number of patents with significant technological or commercial value (Griliches, 1990). Therefore, researchers have long been searching for ways to identify valuable patents, which can then be analyzed in more detail (Harhoff et al., 2003; van Zeebroeck, 2011).
Several studies in recent years have applied connectivity algorithms to the network formed by patents (as vertices) and patent citations (as arcs) in order to identify technologically significant patents (Choi and Park, 2009; Bekkers and Martinelli, 2012; Epicoco et al., 2014; Ho et al., 2014). The idea is that patent citations contain valuable information about knolwedge ‘inheritance’ between patents and can thus be used to identify key linkages in technological evolution (Martinelli and Nomaler, 2014).
119
External validations show that this approach can reduce a large patent dataset down to a small selection of patents that were highly relevant for technological progress at the time of filing (Fontana et al., 2009; Barberá-Tomás et al., 2011). The sequence of these relevant patents is a representation of the core of the technological trajectory and gives insights into how the focus of innovative activity changed as the technology evolved over time (Verspagen, 2007; Martinelli, 2012; Epicoco, 2013). Huenteler et al. (2014b) further demonstrate that the topical focus of patenting along the technological trajectory also corresponds well to trends in innovative activity in the industry and that patent-citation networks therefore can be used to identify the emergence of dominant designs and technology life- cycle patterns. However, until now only few studies have combined this approach with a systematic representation of the technological system and classified the identified patents accordingly, as it has been done in detailed analyses of technological evolution in specific fields (e.g., Rosenkopf and Nerkar, 1999; Prencipe, 2000).
This paper integrates a citation-network analysis with a manual classification of the identified patents. First, the paper develops a patent and patent-citation dataset for solar PV and wind power for the period 1963-2009 (described in detail in section 4.2). Second, we apply two connectivity algorithms to this dataset to identify the core trajectory for both technologies (section 4.3). And third, we group the top 1,500 patents according to their technological focus – e.g., product design versus production process – to identify whether the technological trajectories match with either of the two representations of the technology life-cycle (section 4.4).
4.2. Patent Data
We compiled the database of patent and patent citation data with the objective to obtain a comprehensive dataset of global patenting in the two technologies over the time period 1963 to 2009.32 The patent data was extracted from the proprietary Derwent World Patent Index (DWPI) database, which collects data from 48 patent offices. We chose DWPI because it facilitated the assessment of patent content by providing expert-generated abstracts of all patents (see section 4.4), including translated abstracts for non-English entries in the database.
32 The search was conducted in 2013 but the database was truncated after 2009 to account for the time-lag between patent filing and publication.
120 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
The search string was developed through a two-step procedure. First, we compiled a list of relevant keywords extracted from the innovation literature (a total of 6 experts from the two industries provided feedback on the identified keywords). Then we iteratively curtailed the keyword list by applying it to the initial set of International Patent Classification (IPC) classes listed in the ‘Green Inventory’ of the World Intellectual Property Organization (such as the class ‘wind motors’ F03D) and manually checking random samples for irrelevant patents.33 Second, additional IPC classes were added to the search string based on information on co-filings of relevant patents. Final tests indicated about 6% and 13% false positives as well as about 9% and 14% false negatives for wind power and solar PV, respectively.34 Because connectivity algorithms are robust to false positives, we focused on reducing the error of exclusion when constructing the search filter – partly at the expense of the error of inclusion. Therefore, after retrieving the citation data of all patents (see below), we extended the database in a second iteration to include those 1,000 outside patents that received the most citations from the patents in the database (almost all of these were relevant solar and wind patents).
The citation data was extracted from the DWPI and Thompson Innovation databases, which together cover most of the patent offices’ data. We cleaned the citation data from duplicate citations between different patents in the patent families and excluded circular references.35 One problem that arises when using citation data is that early patents have a disproportionally high likelihood of being getting cited because the population of potential citing patents is higher than for new patents. Therefore, in order to avoid a bias toward older patents, we discarded all citations with a lag between filings of citing and cited patent of more than five years. In a last step we removed all unconnected patents, i.e., all
33 We applied the keywords to the titles, abstracts and claims of patents. 34 To test for false positives, we randomly tested a total of about 1,000 patents for each technology (50 patents for each of the 18 and 20 four-digit IPC classes in the search strings for solar PV and wind, respectively). For false negatives, we checked how many of the patents filed by the top 12 pure-player PV manufacturers (by 2012 cell market share) and 8 pure-player wind turbine manufacturers (in 2010 by market share) were included in our database. 35 Whenever we found circular references, i.e., mutual citations between patents, we deleted the citation coming from the patent with the earlier priority date. Such citations can occur when examiners add citations to new patents filed during the examination process, or when patents are filed in multiple countries.
121
patents without citation link to any other patent in the database. The final database contains 26,775 solar patent families36 (55,687 linkages with a lag ≤5 years) and 8,907 wind patent families (18,718).
Given the time period represented in the database, our analysis is able to reliably identify technologically significant patents until at least 2005. Figure 5 shows how patents and citations are distributed over time. Both technologies saw a first increase in activity around 1980 and then an exponential increase in patenting from about 1995. Received and filed citations increase about proportionally to patent filings until 2005, after which received citations drop rapidly because patents after 2005 did not have a full five-year window of possible citing patents in the database.
a) Solar PV b) Wind power 10.000 5.000 4.000 1.500
Patents (right axis) 8.000 Patents (right axis) 4.000 Citations 3.000 Citations Citations received 1.000 6.000 Citations received 3.000 2.000 4.000 2.000 500 1.000 2.000 1.000
0 0 0 0 19701975 1980 1985 1990 1995 2000 2005 2009 1971 19751980 1985 1990 1995 2000 2005 2009
Figure 5: Descriptive statistics for patent filings, filed citations and received citations over time. Only citations with lag of ≤ 5 years are included. The trends in patenting are in line with other studies who find a surge in patenting activity in the era of incremental change (Gort and Klepper, 1982; Lee and Berente, 2013).
4.3. Connectivity Analysis
In order to identify differences in the development of solar PV and wind power we applied connectivity algorithms to the patent data. We designed the analysis in order to address two aspects of the broader research question: In step I, we identified the current trajectory of innovative activity and traced back the technological foundations of this current trajectory. The results of this step are used to characterize the current stage of the technological lifecycle in the two technologies (i.e., at the end of the observed period in 2009) and can yield insights into where the technology is heading at the moment. In step II, we analyzed how and when the current trajectory emerged as the industry’s dominant trajectory and which alternative paths of development existed in the past (and were
36 We used patent families instead of individual patents to avoid double-counting of multiple filings in different offices.
122 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation abandoned). The results of this step are used to characterize the technology life-cycle as a whole, including significant shifts in the focus of innovative activity in the past. For both analyses, we used connectivity algorithms to extract sub-networks small enough to be categorized manually (see Section 4.4).
Both analyses employ the search path link count (SPLC) algorithm and the critical path method (CPM). The SPLC algorithm aims to identify the most important arcs (i.e., citations) in the network (Hummon and Doreian, 1989; Verspagen, 2007). A ‘search path’ is every possible way from a sink in the network (i.e., a patent that only cites and does not get cited) to a source (patents that only get cited). The ‘link count’ enumerates all possible search paths in the network, and counts how often an arc lies on such a search path. The count is then assigned as weight to each adjacent patent, thus identifying patents along the most important technological linkages in the network. Because the weight of patents in the network is highly skewed, with a few patents holding most of the aggregate weight, this algorithm can be used to reduce the complexity of the network significantly – e.g., in the case of wind power 158 of the 8,907 connected patents hold between them 80% of the total weight (494 patents hold 95%). Building on the results of the SPLC, the CPM determines the search path with the largest total sum of arc weights (e.g., Fontana et al., 2009; Epicoco et al., 2014). We implemented the algorithms using Pajek (de Nooy et al., 2011).
To characterize the current stage of the technological lifecycle (step I), we applied the SPLC and the CPM to the full network 1963-2009 for each technology (networks B in Table 3 below) to identify the core trajectory or ‘backbone’ of the trajectory (sub-networks C in Table 3) (Epicoco, 2013; Prabhakaran et al., 2014). As a robustness test we also extracted and analyzed the top 80% and top 95%-weight networks (applying a so-called ‘vertex-cut’ algorithm; D and E) (Batagelj and Mrvar, 2004). This first step reveals the most important patents and citation linkages in the full network – i.e., the current dominant trajectory and its technological roots. However, it does not reveal when the current trajectory was selected or what the alternatives where. Because the algorithm uses all information contained in the network to evaluate each patent, the evaluation of patents filed in year t change over time as new patents are filed in t+1, t+2, etc. This means that previously important trajectories that turned out to be dead-ends are no longer visible. Therefore, step II is necessary to analyze the technology life-cycle in ‘real time’.
123
To characterize the technology life-cycle as a whole (step II), we applied the CPM to a series of 35
37 gradually growing networks Nt, starting with a network N1975 covering the years 1963-1975 and ending with the full network N2009 covering 1963-2009 (displayed in Figure 10 are 8 of them, in 5-year steps). We then merged the critical paths into one network and color-coded each node by the last network Nt in which it is part of the critical path (sub-networks F in Table 3). This analysis reveals dead-ends and abandoned trajectories hidden in the data. Descriptive statistics of the full networks and all sub-networks are provided in Table 3 below.
Table 3: Descriptive statistics of patent data.
A B C D E F
Sequential critical Technology Full network Critical path 80%-weight network 95%-weight network Full network paths (linkages with (linkages with lag ≤ 5 (linkages with lag ≤ 5 (linkages with lag ≤ 5 (all linkages) (linkages with lag ≤ 5 lag ≤ 5 years) years) years) years) years)
1963-1975 … Time period 1963-2009 1963-2009 1963-2009 1963-2009 1963-2009 1963-2009
32,919 26,775 Solar PV 35 (53) 322 (1,063) 915 (2,069) 3 (2) … 35 (53) (129,993) (55,687)
Wind 11,330 8,907 36 (60) 158 (499) 494 (1,827) 4 (3) … 36 (60) power (41,268) (18,718)
4.4. Patent-Content Analysis
As a final step, we manually coded the abstracts and claims of the patents in the sub-networks C-F extracted in step 3 in order to identify the focus of innovation over the technology life-cycle.
The classification of the patent abstracts was done according to the coding schemes shown in Table 4 (solar PV) and Table 5 (wind power). For each of the two technologies, we differentiated 5 functional elements of the system: The system level (i.e., inventions that claimed entire PV system or wind turbine designs) and four different sub-systems each: in the case of PV systems, we classified patents relating to (1) cells, (2) modules, (3) mounting & tracking systems and (4) grid connections patents; in
37 The year 1975 was chosen as a starting point because at that time the cumulative number of patents exceeded 100 for both technologies (257 for PV, 111 for wind).
124 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation wind turbines, we classified patents relating to (1) rotors, (2) power trains, (3) mounting & encapsulation systems and (4) grid connections. In addition, we classified within each sub-system category (e.g., cells, rotors) whether the patent refers to product innovations or process innovations. Table 4 and Table 5 give examples for each of the resulting 9 classes of patents per technology.
One mechanical engineer and one electrical engineer independently classified each of the patents according to the abstracts’ focus in the technological system. Overall the agreement between the two coders was 87%. In cases of disagreement a consensus was reached after discussing the patent content in detail.
As a last step, we discussed our results for the focus of innovative activity over time with academic experts on the solar PV (5 experts) and wind power industries (4). All nine confirmed the trends displayed in the data.
Table 4: Coding scheme for patents in solar PV.
Content code Content Example
Novel PV system design in which novelty has to do with the design of Tubular photovoltaic solar cells situated at the focus of a PV system at least two of two sub-systems (cell, module, mounting system and line-generated parabolic reflector (US 3,990,914) grid connection)
Layered photovoltaic cell with more than one active Product Novel design of cell or cell materials junction for higher efficiency (US 4,017,332) Cell Production process for crystalline thin-film cell (US Process Novel production process for cell or cell materials 5,130,103)
Novel design of module, including cell separation, cell interconnection Amorphous silicon solar cell element encapsulated by a Product or cell encapsulation, including specific materials and components filler with low moisture permeability (US 5,344,498) Module Novel production process for module, module materials or module Solar cell module manufacturing method with improved Process components sealing characteristics (US 20,040,191,422)
Modular PV mounting system with batten-and-seam Novel design of array, mounting system or tracking system (including Product type interconnection that can be attached to roof (US control system) Mounting 5,232,518) system Novel production or installation process for array, mounting system or Method to install rooftop solar system (US Process tracking system 20,010,034,982)
Novel design of inverter, cabling, storage or control system (incl. grid Circuitry design for PV system with earth leakage circuit Product integration control system) breaker (US 6,107,560) Grid connection Novel manufacturing or installation method for inverter, cabling, Process Inverter manufacturing method (JP 4,915,907) storage or control system
125
Table 5: Coding scheme for patents in wind power.
Content code Content Example
Novel wind-turbine design in which novelty has to do with the Vertical axis turbine with novel rotor and novel drive-train Wind turbine design of at least two sub-systems (rotor, power train, arrangement (US 3,902,072) or horizontal-axis rotor with system mounting & encapsulation, and/or grid connection) rotor-integrated generator (US 4,289,970)
Novel design of rotor or rotor components (incl. rotor control Rotor arrangement with teetering hub and rotor control Product system) mechanism (US 4,201,514) Rotor Novel manufacturing or installation method for rotor or rotor Process Rotor blade manufacturing method (JP 4,641,366) components
Novel design of power train or power train components (incl. Product Compact, gearless power train (US 6,921,243) power train control system) Power train Novel manufacturing or installation method for power train or Manufacturing method for magnets of multi-polar Process power train components generator (EP 2,389,512)
Novel design of nacelle, tower or foundation (incl. climate and Tower-nacelle arrangement in which transformer is Product vibration control system) mounted inside the top of the tower (US 7,119,453) Mounting & encapsulation Novel manufacturing or installation method for nacelle, tower Installation method for offshore wind turbine tower (GB Process or foundation 2,460,172)
Electrical connection of wind turbines in a wind farm, Novel design of transformer, substation, cabling or wind farm Product including substation and individual transformers and cabling integration (incl. grid integration control system) (US 7,071,579) Grid connection
Novel manufacturing or installation method for transformer, Process Method of mounting power cables (ES 2,283,192) substation or cabling
5. Results
This results section is structured in line with the sequence of analyses presented in the methodology section. We start by characterizing the current stage of the technology life-cycle of the two technologies (section 5.1; analysis step I of the connectivity analysis). Then we characterize the technology life-cycle as a whole, including significant shifts in the focus of innovative activity in the past (section 5.2; analysis step II). The results reveal strongly contrasting development paths in the two industries. In the development of solar PV technology early product innovations were soon followed by a series of interlinked product and process innovations. In the current stage of the life-cycle, most innovative activity focuses on the cell production processes. In wind turbine technology, in contrast, the focus of innovative activity shifted over time through different parts of the product, rather than from product to process innovations, and is now on incremental product improvements in interdependent sub-systems and components .The consequences of these striking differences for policy makers are discussed in section 6.
126 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
5.1. Characterizing the Current Life-Cycle Stage
The core trajectories in the full networks of solar PV and wind power (Figure 6a and b) allow us to characterize the current stage of the life-cycle, including the technological foundations of current innovative activity. Two main differences between the technologies stand out. First, the breadth of innovative activity is remarkably different: the critical path in PV remains focused on the cell, with only two module patents as exceptions, whereas innovative activity in wind power is spread much more evenly across the four sub-systems: 8, 10, 15 and 3 patents in the rotor, power train, grid connection and mounting & encapsulation, respectively. Additionally, the path in the wind network shows a sequential pattern, focusing first on the rotor (which can be seen as core sub-system), until 1987, before shifting to the power train (mid-1980s to mid-2000s), grid-connection issues (from late 1990s) and mounting & encapsulation structures (since the early 2000s). Second, the two technologies differ in the type of innovation along the trajectory, in particular the relative emphasis on product and process innovations. As can be seen from the color coding in Figure 6a, the current innovative activity in solar PV is almost exclusively focused on the cell production process. Indeed, 25 of the last 26 nodes on the critical path, covering the period 1987-2009, are cell process innovations. Only the first 9 patents and one later patent (in 2004) on the critical path are product innovations. The wind network in Figure 6b, in contrast, shows virtually the opposite: There is not a single process-related patent on the critical path; in fact only 3 of the top 494 patents representing the top 95% of the vertex weight (network E) relate to the production or installation process. More detail on the patents on the critical paths is presented in Table A3 and Table A4 in the appendix.
a) Solar PV b) Wind power PV system Wind turbine architecture architecture
Cell Rotor
Module Power train
Mounting system Grid connection
Mounting & Grid connection encapsulation
Sequence of patents along critical path Sequence of patents along critical path
Patent relating to product innovation (size ~ node weight) Citation ≤ 5 year lag (width ~ link weight) Legend: Patent relating to proces innovation Citation > 5 year lag
Figure 6: Critical path in full networks (network C in Table 3) showing the currently dominant trajectory of innovative activity.
127
The patterns observed in Figure 6 allow to draw conclusions about the innovation process in the era of incremental change in the two technologies: in solar PV, the current trajectory of innovative activity is dominated by cell process innovations, which draw relatively little on knowledge developed for other parts of the system (such as mounting structures or grid integration routines). In contrast, the current trajectory of innovative activity in wind power is centered on product innovations. The last patents on the critical path focus on vibration control in the tower. These product innovations draw not only on knowledge from the sub-system in question but are based also on innovations in other parts of the system, as can be seen from the citations that cross sub-system boundaries (citations with a lag of more than 5 years were not included when analyzing the connectivity, but are nonetheless shown in Figure 6 to illustrate the multitude of linkages between patents in wind power). This points toward the complexity of the product architecture and the ‘systemic’ nature of innovation in wind power.
The two observations from the critical paths remain valid when looking at quantitative indicators describing the broader trajectory, shown in Figure 7. (These analyses are based on the 80%-weight networks D, which are also shown as graphs in Figure A1 in the appendix). Figure 7a and b show comparable data for the breadth of innovative activity, represented by the share of innovative activity in different parts of the system for solar PV and wind power. The graphs illustrate that the focus on the cell sub-systems remains more or less unchanged (cell innovations represent between 60% and 90% of the weighted activity for most of the observed period). Opposed to that, the focus in wind turbine technology is sequential and shifts through different parts of the system in a way that each sub- component has a share of at least 40% of the weighted activity in different time periods. The type of innovation can be compared in Figure 8a and b. In solar PV the focus shifts over time from product innovations, which represent an average of 64% of the weight between 1972 and 1985, to process innovations with an average 73% of the weight in 1990-2009. The focus of innovative activity in wind power did not shift to process innovations (of which there are none in the 80%-weight network), but to systemic patents, as shown in Figure 8b. Systemic patents are defined here as patents that received more than half of their citations from patents in other sub-systems (the 7 system-level patents were excluded from this analysis). Their share increased from 25% in 1975-1979 to 63% in 1990-94 and 58% in 2005-09. This, again, illustrates the systemic nature of innovation in wind power, as did the patterns of citations seen in Figure 6b.
128 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
a) Solar PV b) Wind power 100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0% 1975 1980 1985 1990 1995 2000 2005 2009 1975 1980 1985 1990 1995 2000 2005 2009
System architecture Mounting system System architecture Grid connection Cell Grid connection Rotor Mounting & encapsulation Module Power train
Figure 7: Share of innovative activity in different parts of the technological system (based on patent-content categorization of 80%-weight networks D).
a) Solar PV b) Wind power 100% 100%
80% 80%
Systemic patents 60% 60% Series 40% 40%
20% 20%
0% 0% 1975 1980 1985 1990 1995 2000 2005 2009 1975 1980 1985 1990 1995 2000 2005 2010
Product patents Process patents
Figure 8: a) Shift from product to process innovation along life-cycle in solar PV, b) Share of ‘systemic patents’ in wind power over time, defined as patents that received more (>50%) citations from patents in other sub-systems than from its ‘own’ sub-system (system-level patents were excluded from this analysis).
5.2. Characterizing Previous Stages of the Technology Life-Cycle
As discussed in section 4.3, the results presented so far allow us to characterize the current stage of the technology life-cycle, but they offer only limited information on shifts in the patterns of innovation in the two technologies in the past. The observation that the later stage of the life-cycle in PV is focused on cell process innovations does not mean that module or grid connection innovations were never important. This section reports results that aim to identify and characterize these past life-cycle stages. The algorithms are the same as those used for the analyses in section 5.1, but were applied not to the full network but to a series of gradually growing networks Nt (where t is the year up to which patents are included in the network).
129
The results for the series of networks yield a detailed picture of how the current trajectories in the two technologies emerged over time, and which alternative trajectories were abandoned. The first main set of observations is contained in Figure 9, which shows the gradual stabilization of the critical paths in the two networks. Specifically, the figure presents a ‘hazard rate’, which is a measure of variation of the core trajectory, for patents on the critical paths of the gradually growing networks (Huenteler et al., 2014b). This hazard rate is to be interpreted as follows: for each year t (on the x-axis), the graph shows how many patents on the critical path of Nt are no longer on the critical path when five years of additional patent data are added to the network – i.e., on the critical path of Nt+5. The decline of the hazard rate in both technologies means that the critical path gradually stabilized over time, although with a major discontinuity in solar PV around 1995 (more below). One can derive from these graphs an approximation of the time when the period of major competition between alternative trajectories ended. This provides insights about the technology life-cycle as a whole, and specifically about the emergence of a dominant design: If one defines a trajectory as stable as soon as it conserves at least 50% of the patents on the critical path over a period of five years (i.e., the hazard rate remains below 50%), a stable technological trajectory emerged in PV in 1996 and in wind power in 1984 (or 1989, when the value is exactly 50%). These dates roughly match with the data on design competition in the market presented in Figure 3 as well as with qualitative accounts of the emergence of dominant designs in the two technologies (Menanteau, 2000; Bergek and Jacobsson, 2003; Nemet, 2009).
a) Solar PV b) Wind power 1,00 1,00
0,75 0,75 trajectory No dominant
0,50 0,50
0,25 0,25 trajectory 0,00 0,00 One dominant 19751980 1985 1990 1995 2000 2005 1975 1980 1985 1990 1995 2000 2005
t t
Figure 9: Hazard rates of patents on the critical path, indicating share of patent that is still on critical path after five years of new patent filings have been added to the network.
130 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
The second set of more detailed observations is contained in Figure 10, which integrates the critical
38 paths of 8 different networks (N1975, N1980, N1985 … N2009) in one graph. Each patent in the graph is colored with a different shade of grey, which indicates the year of the last critical path the patent is part of. This figure allows us to identify two aspects of the earlier stages of the technology life-cycle. First, it shows how the focus of innovative activity in the two technologies evolved in ‘real time’, because unlike in Figure 6 the evaluation of earlier patents is not influenced by the (ex-post) information which trajectory eventually ‘succeeded’. When comparing the graphs to those shown in Figure 6 above, it becomes clear that competing trajectories, indicated by the gray shades and the non- white patents in the graph, existed in PV mainly until around 1995. In wind power the currently dominant trajectory had already emerged by the late 1970s – only a few non-white patents are located on alternative trajectories that branch off here and there in the late 1970s and mid-1980s. In terms of solar PV, the graph suggests that the industry already focused strongly on production processes in early stages. However, opposed to what can be observed when looking at the currently dominant trajectory in Figure 6a, there was also a period (until 1995, and then again briefly in 2002-03) when module product and process innovations were very important. In wind power, the graph demonstrates that the current trajectory has been dominant for so long (compare Figure 9) that the additional critical paths add little information to the analysis of the focus of innovative activity. Additionally, it is noteworthy that not a single patent on any of the eight critical paths has been on the process level which reinforces the observation made from Figure 6b.
Second, Figure 10 allows us to identify trajectories that had been important but are now out of focus. The graph contains three such trajectories in each of the two technologies. In solar PV, all three exhibit a stronger focus on the module subsystem than the current trajectory. The first two trajectories, which end in 1980 and 1995 and are marked (a) and (b) in the graph, both show a pattern of close linkage between product and process innovations. The first set of these, marked (a), focuses on ways to encapsulate solar cells in laminates that are radiation-transparent and protect the cells from water and other environmental influences (e.g., US 4,067,764, US 4,009,054, US 4,224,081). These innovations
38 To test the robustness of this approach, we compared the network combing the 8 critical paths (network ‘I’) to one that combines all patents that are on at least three critical paths (‘II’). In solar PV, all 65 patents of II are also part of I, which contains 92 patents. In wind power, II contains 50 patents, 38 of which are part of I, which has 47 patents; those that are not on I are patents from the late 1970s and early 1980s on the system level and in the sub-system rotor, thus adding little information to Figure 10.
131
are technologically independent of the current trajectory but are nonetheless still important parts of current modules. The second set of patents (b), which spans a period from the late 1970s to the mid- 1990s, relates to the electrical integration of thin-film modules (e.g., US 4,315,096, US 4,624,045and US 4,650,524), a technology that was long regarded as the most promising technology but which is now increasingly marginalized (see Figure 3 above). In thin-film solar PV, the process of module and cell manufacturing is much more integrated than in crystalline silicon PV, which is reflected in the stronger focus on module patents on this trajectory.39 The closer integration of product and process innovations in thin-film PV, too, reflects characteristics of the technology, for two reasons. On the one hand, there are a lot more design variations possible due to a larger choice of possible materials. On the other hand, the economic and technological feasibility of alternative thin-film cell designs and materials hinges almost entirely on the production process, because the production process (i) is even more automated than that of crystalline-silicon cells and (ii) does not allow using production equipment from the chip industry.40 This close relationship suggests that the predominant focus on cells and the production process might today be slightly different if thin-film modules had emerged as dominant trajectory. A third, more recent abandoned trajectory contains patents relating to encapsulation and mounting elements (c) as well as patents relating to the production of specific materials for thin-film cells (d). The latter suggest that renewed focus on thin-film cells in the mid- to late 2000s in some parts of the industry (cf., Figure 3) is also reflected in the patent network.
39 Thin-film modules are produced by depositing a thin film of semiconductor material on a large surface, cutting individual cells into the large surface and then connecting these cells electronically. That means cell and module manufacturing are part of one integrated production process, which is why they are almost never physically or organizationally separated. In crystalline silicon, in contrast, cells are produced on small surfaces first and then integrated into a module. 40 See, e.g., Jager-Waldau (2004). Indeed, manufacturers of thin-film modules had and still have much more problems to translate the high-efficiencies and high-yields of smaller, laboratory-constructed cells to production volumes (e.g., Razykov et al., 2011).
132 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
a) Solar PV
PV system architecture
Product Cell Process (d)
Product Module Process (a) (b)
(c) Mounting system Product Process
Product Grid connection Process
Patent on critical path of full network (-2009) b) Wind power -2005 -1995 -1985 -1975 -2000 -1990 -1980 Wind turbine architecture
(c) Product Rotor Process
Product Power train Process
Grid Product connection (a) Process
Mounting & Product encapsulation (b) Process
Figure 10: Network for solar PV and wind power which combine patents from the 8 critical paths of networks N1975, N1980,
N1985…. N2009 to illustrate competing trajectories and emergence of currently dominant trajectory. The color of each patent (node) indicate the year of the last critical path that the patent is part of. The letters (a)-(c) in solar PV and (a)-(b) in wind power indicate ‘abandoned’ trajectories.
The wind power graph shows three alternative trajectories that branch of early on and are representative of alternative technological paths pursued in the early days of the wind industry. The first one, marked (a), is representative of a few early critical paths that focus on alternative, vertical-axis rotor designs (e.g., US 3,883,750, US 4,012,163, US 4,115,027), a technological path that was pursued in the 1970s and 80s but then quickly abandoned outside of small niche applications. Connected to this is the option to store electricity in a flywheel, which can be linked to vertical axes turbines more easily than to current turbines (US 4,171,491, US 3,944,840, US 4,035,658). The one marked (b) branches off to an early patent claiming a mechanical mechanism to control vibrations induced by the
133
reorientation of a horizontal rotor to changing wind directions (US 4,692,094; also US 4,557,666). In the late 1980s several further critical path patents are linked to alternative, mechanical mechanisms to control the rotational speed of a rotor of a horizontal axis turbine (e.g., US 5,096,378, US 4,692,095), such as the one trajectory marked by (c) in Figure 10 which branches off to a mechanical rotor control system (using a spring and a rotating mass which adjusts the orientation of each blade to the wind to avoid over-speeding). These represent alternatives to electronic control systems, which is now standard throughout the industry.
6. Discussion
Our results suggest that solar PV and wind power followed very different technology life-cycles over the last four decades, but that both patterns can be explained with existing theoretical models. Linking the temporal patterns in solar PV and wind power to the theoretical models allows us to draw conclusions from the literature about the learning and innovation processes in the two technologies. The models point toward very different innovation and learning processes in the two technologies – differences that are likely to be even wider when looking at the entire technology space in the energy sector, as discussed in section 6.1. The different innovation and learning processes imply the need to tailor technology policy to technological characteristics (6.2). The findings further help conceptualizing previously inconclusive evidence about the impact of technology policies in the past (6.3).
6.1. Technology Life-Cycles in Energy Technologies
Our results demonstrate that the technology life-cycle of solar PV conforms well to the predictions of the A-U model of mass-produced goods: early product innovations were followed by a surge of process innovations in solar cell production. Wind power, on the other hand, went through a life-cycle that closely resembles the predictions from the Davies model for the life-cycle of complex-products and systems: after an initial period with competing product architectures, the focus of innovative activity shifted over time through different parts of the product, rather than from product to process innovations.
As discussed in section 3.1, the two technologies differ in the two main determinants of these patterns, the complexity of the product architecture and the scale of the production process. However, they are by far not the most extreme cases within the energy sector. When going beyond the technologies
134 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation analyzed in this paper, it quickly becomes clear that the dichotomy of ‘complex products and systems’ and ‘mass produced technologies’ alone does not suffice to describe the full variety of energy technologies. Figure 17 locates a broader set of energy technologies in the technology space generated by the two characteristics. Complex products and systems can be further divided into infrastructure systems, such as public transport systems and electricity grids, and design-intensive products, which include most large electric power plants. Infrastructure systems are highly complex and provided through a project-based production process, and thus involve hardly any process innovation. Design- intensive products are manufactured in small but significant quantities and thus involve some form of process innovation. On the other end of the spectrum, mass-produced goods are divided into continuous-flow processes (such as the production of transport fuels), for which the process is the primary focus of innovation from beginning, and process-intensive products, which involve some experimentation with different product designs in the beginning (solar PV, fuel cells. The graphic also shows two groups of technologies that do not fit on the continuum between mass-produced goods and complex products and systems: (i) low-tech products (small wind, small hydro turbines) which are relatively simple and produced in very small batches, and have potential for neither significant product nor process innovation, and (ii) mass-produced complex products (electric cars, grid-scale batteries), which involve continued product and process innovations over the entire technology life-cycle.
Infrastructure systems Transport systems, High electricity grids
Design-intensive products Mass-produced complex products Complexity of product Gas turbines, wind Electric cars, grid-scale batteries architecture Low-tech products Process-intensive products Small hydro, small wind Solar PV, fuel cells
Low Continuous-flow processes Biofuels, building materials
Low High Scale of production process
Figure 11: Stylized classification of different energy technologies according to scale of production process and complexity of product architecture.
When comparing the case technologies with those listed in Figure 11, it becomes clear that solar PV and wind power are in fact relatively similar. Wind turbines can be positioned in the lower right corner of complex products and systems, which implies that the systemic nature of innovation will be even
135
more pronounced in these cases than observed in wind power, so will the need for learning-by-using and user-producer interaction. Solar PV systems can be positioned in the upper left corner of mass- produced goods, some of which will thus be even more pronounced in the focus on process innovations, which also implies more pronounced economies of scale and learning-by-doing potentials. The technologies that are located to the left or right of the diagonal in Figure 11 are more difficult to characterize. Deducting from the patterns observed for the technologies on the diagonal, low-tech products can be expected to have relatively little absolute potential for learning and cost reductions; mass-produced complex products, on the other hand, can be expected to exhibit large potentials in both areas of learning and economies of scale.
6.2. Implications for Technology Policy
The life-cycle patterns identified in this paper point toward very different sources of relevant experience and potentials for innovation in the two analyzed technologies and the energy technology space in general. This section explores the implications of these differences for technology policy. A particular focus is on the design of so-called ‘deployment policies’, because in recent years, rather than focusing purely on public investment in R&D, many countries provide public resources for the deployment of clean technologies in order to induce innovations and ‘buy-down’ cost (e.g., PCAST, 1999; Gallagher, 2014).41 Much of the policy debate on the function of such deployment policies in the innovation process is centered on learning-by-doing in manufacturing and economies of scale, reflecting the A-U technology life-cycle model. However, our analysis shows that the energy sector comprises technologies that do not conform to this model of the technology life-cycle.
Deployment policies typically target relatively mature technologies. The two contrasting models of the technology life-cycle discussed in section 2.1 suggest that technological trajectories in the energy sector differ in three aspects that affect the role of deployment – and thus, the potential role of deployment policies – in the innovation process in later stages of the technology life-cycle: First, economies of scale in manufacturing, and thus the absolute size of the supported market, are much more important for mass- produced goods than for complex products and systems. Mass-produced goods need the prospect of a large market to realize economies of scale in manufacturing and to justify investments into R&D for
41Overall, the International Energy Agency estimates that the world will spend up $1.2trn on wind power and $1.1trn on solar PV over the next 25 years (IEA, 2014).
136 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation specialized production equipment and materials. If the prospect of such a market is too uncertain, a ‘chicken-and-egg’ situation can arise in which the market does not grow because cost are too high and cost cannot come down because the market is too small (e.g., Cantono and Silverberg, 2009). In complex products and systems, where most production facilities remain general-purpose, other variables besides market size are more important for the empirical relationship between deployment policies and innovations or cost reductions. Second, by facilitating feedback cycles between R&D and technology users, deployment can play a significant role in reducing technological uncertainty in complex products and systems, where uncertainty about product performance and user needs remain high throughout the technology life-cycle. While existent, the benefits from additional long-term and large-scale testing for the R&D process can be expected to be much smaller in mass-product products. Third, because user-producer interaction is so important, geographical and organizational proximity of markets and users can be very important for the R&D and innovation process in complex products and systems. In contrast, proximity appears much less relevant for mass-produced goods.
These three characteristics can serve as guideposts for technology policies that aim to make use of deployment to stimulate innovation (see Figure 12 and Table 6). For mass-produced goods, large markets, ideally coordinated internationally, are needed to enable the necessary economies of scale and the learning-by-doing in production. At the same time, policy support needs to make sure that cost competition remains high, e.g., by auctioning off subsidies, by dynamically adjusting incentives or by requiring supported firms to publish cost data. For larger and more complex technologies such as wind turbines, geothermal systems, nuclear power plants, and tidal energy systems, deployment policies have to go beyond simply subsidizing scale to in order to fully realize their potential innovation impact. For these technologies, deployment policies need to be understood as R&D policies rather than merely as subsidies. Rather than enabling economies of scale, deployment policies should be targeted at creating ‘performance-driven’ niche markets (Grubler and Wilson, 2014): they should not aim for very large roll-out of existing technologies, but be explicitly be targeted at reducing technological uncertainty, for example by providing grants to innovative product features, tying subsidies to requirements to publish cost and performance data, or by financing experimentation in different geographical and climatic environments. Furthermore, deployment policies could be accompanied by measures to enhance user- producer interaction (e.g., technology platforms or grants for consortia), improve market transparency (through collecting and publishing performance data) and gradually adjust performance standards (e.g., as it has been done with grid-integration requirements for wind turbines).
137
Experimentation and user-producer interaction Grants for innovative features High Consortia, private-public partnerships Domestic markets Performance competition Complexity of product Data collection and publication architecture Evolving standards Scale Large, ideally global markets Cost competition Low Reverse auctions Rapid adjustment of incentives
Low High Scale of production process
Figure 12: Characteristics of deployment policies if tailored to the characteristics of the two life-cycle models.
Mass-produced energy technologies Complex energy technologies
Enabling economies of scale & learning by doing Enable full-scale experimentation in use- Primary objective in commercial-scale production processes, enable environment, reduce uncertainty about product manufacturer-supplier interaction innovations, enable user-producer interaction
Geographical scope Large-scale (ideally global) Close to producers
Manufacturers & their suppliers (materials, Primary actors in innovation process Users, manufacturers and component suppliers production equipment)
Evolving requirements and technological Cost competition drives innovation -> opportunities drive innovation -> incentivize governments need to continuously adapt Creating pressure to innovate continuous experimentation; create transparency remuneration, minimize entry barriers and about performance characteristics; monitor and standardize regulation across jurisdictions continuously adapt performance requirements
Grants for innovative features; consortia; private- Complementary policies Rapid adjustment of incentives, reverse auctions public partnerships
Table 6: Characteristics of deployment policies tailored to the characteristics of the two life-cycle models.
6.3. Reconciling Empirical Evidence
Our analysis provides quantitative evidence for systematic differences between solar PV and wind power. This evidence helps reconciling two areas of conflicting evidence about the impact of technology policies on innovation.
First, there is an ongoing academic debate about whether subsidies for technology deployment can stimulate innovation and technological learning, or just enable firms to exploit existing designs and economies of scale (Nemet, 2006, 2009; Hoppmann et al., 2013). The life-cycle models that match our findings for the two technologies suggest that the effect depends on characteristics of the supported technology. Indeed, deployment subsidies in solar PV primarily enabled innovations in
138 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation manufacturing (Norberg-Bohm, 2000; Hoppmann et al., 2013) and cost reductions through economies of scale (Nemet, 2006). In wind power, in contrast, experience generated in government- supported markets was a key driver of product innovation (Andersen, 2004). However, a very large market alone was not sufficient to stimulate innovation in wind turbines, as experience with the early US wind policies suggests (Nemet, 2009). Rather, deployment subsidies in wind power worked best when they were combined with measures to facilitate learning by interacting in the form of knowledge transfer between turbine producers, turbine owners and researchers (Kamp, 2004; Tang and Popp, 2013).
Second, our analysis also provides a starting point to explain the importance of ‘home markets’ for technological innovation which has been observed for some energy technologies but not for others. The technology life-cycle patterns revealed in this study suggest that geographical proximity to users remains important for innovators in complex technologies such as wind power, while it is no longer required in a technology like today’s solar PV. These predictions match very well with the empirical evidence. Two recent, analogous econometric studies analyzed the effect of deployment policies on domestic and foreign innovation in wind power and solar PV. Dechezleprêtre and Glachant (2013) find that domestic wind power deployment policies had an effect on innovation 28 times stronger than foreign ones. In contrast, while there is evidence for such a relationship in the early days of the industry (Hoffmann et al., 2004), Peters et al. (2012) do not find a significant difference between the effects of domestic and foreign deployment policies in solar PV. A similar picture emerges from studies analyzing the effect of deployment policies on the competitive success of domestic firms. On the one hand, comparative studies of wind power in different countries find that domestic deployment policies correlate well with industrial competitiveness.42 Lewis and Wiser (2007) conclude from a review of global wind power industry development that domestic deployment policies are “a pre- requisite to achieving successful localization” (p. 1855; italics added). On the other hand, recent quantitative studies of the PV industry find that domestic market size is not a good predictor of trade competitiveness (ICTSD, 2010; Algieri et al., 2011). Similarly, recent reports by policy think tanks explicitly compare deployment policy outcomes in the solar PV and wind power industries (Huberty and Zachmann, 2011; Barua et al., 2012). Using trade data, Huberty and Zachmann (2011) find a
42 The market leaders of the four largest markets in 2010 – China, the US, India, and Germany – were all domestic companies (BTM, 2011).
139
correlation between domestic deployment and competitiveness in wind power, but no such relationship in solar PV. They arrive at the conclusion that using domestic demand as an industrial policy “may work for wind turbines, but we find no evidence that it works for solar cells” (p. 1). Similarly, Barua et al. (2012) conclude from a multi-country case study that “domestic deployment is key to building ... domestic industries” in wind power, whereas in PV “a large domestic manufacturing industry and significant domestic deployment do not necessarily go hand-in-hand” (p. 2-3).43 The differing role of geographical proximity is reflected in processes of catching up of emerging economies in the two industries. In wind power, catching up almost always involves significant support for a domestic market, and often required protectionist actions by governments (Lewis, 2007, 2011). The cases of China, Taiwan, and Malaysia, in contrast, which emerged as hubs of PV cell and module production without supporting a significant domestic market, show that countries can reach competitiveness in PV manufacturing without supporting local demand (Liu and Goldstein, 2012; Cao and Groba, 2013).
6.4. Limitations and Further Research
An empirical study as the one presented in this paper has several inherent limitations. Since the validity of the implications formulated above for the design of technology policy hinges on the validity of the applied methodology, three aspects have to be highlighted, which lend themselves as avenues for future research. First, using patents as indicators for innovation introduces a bias against process innovations. Since much the relevant information is to be revealed anyway, a product innovation is more likely to be patented than a process innovation, for which inventors may choose other means of appropriability, most notably secrecy. For example, Arundel and Kabla (1998) find that the European machinery firms patented about 52% of product innovations and 16% of process innovations. The fact that we found very few process patents in wind power along the trajectory may be due to a bias against process knowledge in general. This makes careful interpretation of results necessary. However, because this bias should be similar for both technologies, it should not affect the conclusion that there are significant differences between the two technologies. Future research could focus on a combination of indicators to assess life-cycle patterns. Second, for lack of available citation data, we could not include Chinese patents in our analysis. From a latecomer position China has caught up quickly in clean
43 In 2011, the top five wind markets (according to cumulative installed capacity) were home to 9 of the top 10 turbine suppliers, whereas in PV the top 5 countries were home to only three.
140 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation technologies since the early to mid-2000s. Especially in solar PV, Chinese firms have come to dominate the global market. Our patent data shows a surge of Chinese patent filings in both technologies since about 2010. Understanding the Chinese firms’ influence on the technological trajectory and the observed life-cycle patterns is highly relevant for the academic literature and the policy community. Once Chinese citation data is systematically available in commercial patent databases, future research should aim to address these questions. Third, our broader conclusions need to be validated by characterizing the life-cycles of additional technologies in the energy sector. The fact that the two selected technologies already show significantly different life-cycle patterns suggests that there is much to learn when comparing the more extreme areas of the space mapped in Figure 11. Especially in the lower left and upper right corners of the framework, intuition suggests that empirical analyses could reveal patterns that have so far not been described by the two traditional life-cycle models. Beyond the energy sector, we believe that the methodology and indicators developed in this paper open up promising research opportunities in toward a systematic characterization of life-cycle patterns across a wide range of technologies.
7. Conclusion
Technological change in energy technology can play a major role in mitigating climate change and reducing the environmental footprint of energy production and consumption. To stimulate the necessary innovation, governments will likely spend trillions of USD of public resources on technology policies for clean energy technologies over the coming decades. This paper mapped the patterns of innovation over the technology life-cycle in solar PV and wind power in order to gain insights about how these resources can be spent effectively.
In particular, the paper analyzed which of two common models of innovation over the technology life- cycle best describes the pattern of innovation in the two technologies. The results suggest that solar PV technology followed the life-cycle pattern of mass-produced goods, a model that typically applies to technologies with relatively simple product architecture and a large-scale production process: early product innovations were followed by a surge of process innovations, especially in solar cell production. Wind power systems, in contrast, more closely resembled the life-cycle of complex products and systems, a model that has been developed for technologies with a complex product architecture and low-volume production: the focus of innovative activity shifted over time from the system architecture and core
141
components to different sub-systems and components of the product, rather than from product to process innovations.
The findings allow to draw conclusions about the patterns of technological learning in energy technologies from the general literature on technology life-cycles, and to make sense of seemingly conflicting evidence about innovation and policy impacts in the two technologies. In solar PV, most innovations after the first large-scale deployment of the technology in the 1980s were focused on the production process, which points toward a predominant role of learning-by-doing, economies of scale in manufacturing and innovations in production equipment. In wind power most innovations introduced novel sub-system and component designs, which points toward the importance of learning-by-using, product up-scaling and innovations in operation & maintenance (O&M). These differing patterns correspond well with existing studies of technological learning in the two technologies and help putting these studies in comparative context.
Besides the conclusions about the innovation process, the contrasting characterizations of the learning processes in the two technologies have important policy implications, in particular with regard to public policies that subsidize and facilitate large-scale deployment and use of these technologies. The different life-cycle patterns suggest that deployment policies play very different roles in innovation in the two technologies: in a learning process that is centered around the production process, deployment policy support can be crucial to enable learning-by-doing, large-scale production and markets for production equipment; in contrast; in a learning process that is centered around the product design, deployment policy support can be crucial to enable learning-by-using, gradual up-scaling and markets for specialized O&M service providers.
Differing roles of large-scale deployment in the innovation process imply different, technology-specific policy instrument designs. These stand in contrast to the current practice of one-size-fits-all instruments that some governments employ to stimulate energy innovation, e.g., through tax credits or feed-in tariffs for all types of renewable electricity, or uniform mandates for all kinds of alternative vehicle drive-trains. For mass-produced goods, such as solar cells, biofuels, LEDs, batteries or fuel cells, large markets, ideally coordinated internationally, are needed to enable the necessary economies of scale and the learning-by-doing in production – a small market, even if supported over a long time frame, will not overcome the ‘chicken-and-egg’ problem of low production volumes and high production costs. For complex products and systems, such as wind turbines, geothermal systems, nuclear power plants, and transport systems, deployment policies have to go beyond simply subsidizing more-
142 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation of-the-same in order to fully realize their potential innovation impact. For these technologies, deployment policies should take the form of ‘performance-driven niche markets’, because these policies are most useful if they generate valuable experience from learning-by-using and can enable user- producer interaction, not if they only enable economies of scale and learning-by-doing.
In conclusion, few people would support a 'one-size-fits-all' innovation policy approach for semiconductor, machinery, biotechnology, oil and gas, and chemical industries. The findings of this paper indicate that it may be equally misleading to lump together solar PV systems, wind turbines, biomass gasification, carbon capture and storage, and fuel cells when designing policy instruments to stimulate innovation in clean energy technologies.
Acknowledgements
Previous versions of this paper have been presented at the Energy Policy Consortium Seminar at Harvard in February 2014, the ECN/ETH Zurich side event at UNFCCC COP 18 in Doha, Qatar, the International Sustainability Transitions 2012 conference in Copenhagen and the Denmark and the International Schumpeter Society Conference 2012 in Brisbane, Australia. We are grateful for the feedback received by the conference and workshop participants. All errors remain our own.
143
References
Abernathy, W.J., Utterback, J.M., 1978. Patterns of industrial innovation. Technology Review 80, 40–47.
Abernathy, W.J., Utterback, J.M., 1988. Innovation over time and in historical context, in: Tushman, M.L., Moore, W.L. (Eds), Readings in the Management of Innovation. Harper Collins Publishers, New York, NY, pp. 25–36.
Algieri, B., Aquino, A., Succurro, M., 2011. Going “green”: trade specialisation dynamics in the solar photovoltaic sector. Energy Policy 39, 7275–7283.
Anadon, L. D., 2012. Missions-oriented RD&D institutions in energy between 2000 and 2010: A comparative analysis of China, the United Kingdom, and the United States. Research Policy 41, 1742–1756.
Andersen, P.D., 2004. Sources of experience? theoretical considerations and empirical observations from Danish wind energy technology. International Journal of Energy Technology and Policy 2, 33–51.
Anderson, P., Tushman, M.L., 1990. Technological Discontinuities and Dominant Designs: A Cyclical Model of Technological Change. Administrative Science Quarterly 35, 604–633.
Arundel, A. V, Kabla, I., 1998. What percentage of innovations are patented? Empirical estimates for European firms. Research Policy 27, 127–141.
Baldwin, C., Maccormack, A., Rusnak, J., 2014. Hidden Structure: Using Network Methods to Map Product Architecture. Research Policy 43, 1381–1397.
Barberá-Tomás, D., Jiménez-Sáez, F., Castelló-Molina, I., 2011. Mapping the importance of the real world: The validity of connectivity analysis of patent citations networks. Research Policy 40, 473–486.
Barua, P., Tawney, L., Weischer, L., 2012. Delivering on the Clean Energy Economy: The Role of Policy in Developing Successful Domestic Solar and Wind Industries World Resources Institute (WRI) Working Paper. World Resources Institute (WRI) and Open Climate Netowrk (OCN) Working Paper, Washington, DC.
Batagelj, V., Mrvar, A., 2004. Pajek—analysis and visualization of large networks. Springer, Berlin, Germany.
Bekkers, R., Martinelli, A., 2012. Knowledge positions in high-tech markets: Trajectories, standards, strategies and true innovators. Technological Forecasting and Social Change 79, 1192–1216.
Bergek, A., Jacobsson, S., 2003. The emergence of a growth industry: a comparative analysis of the German, Dutch and Swedish wind turbine industries, in: Cantner, U., Metcalfe, J.S. (Eds), Change, Transformation and Development. Physica, New York, NY and Heidelberg, Germany.
144 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
BNEF, 2012a. Solar Power Update 2012. Bloomberg New Energy Finance (BNEF), New York, NY.
BNEF, 2012b. Wind Energy Innovation. Bloomberg New Energy Finance (BNEF), New York, NY.
Bolinger, M., Wiser, R., 2012. Understanding wind turbine price trends in the US over the past decade. Energy Policy 42, 628–641.
Brusoni, S., Prencipe, A., Pavitt, K., 2001. Knowledge Organizational Coupling, and the Boundaries of the Firm: Why Do Firms Know More Than They Make? Administrative Science Quarterly 46, 597–621.
BTM, 2011. World Market Update 2010. Ringkøbing, Denmark.
Campbell, D.T., 1990. Levels of organization, downward causation, and the selection-theory approach to evolutionary epistemology, in: Greenberg, G., Tobach, E. (Eds), Theories of the Evolution of Knowing. Lawrence Erlbaum, Hillsdale, NJ, pp. 1–17.
Candelise, C., Winskel, M., Gross, R.J.K., 2013. The dynamics of solar PV costs and prices as a challenge for technology forecasting. Renewable and Sustainable Energy Reviews 26, 96–107.
Cantono, S., Silverberg, G., 2009. A percolation model of eco-innovation diffusion: The relationship between diffusion, learning economies and subsidies. Technological Forecasting and Social Change 76, 487–496.
Cao, J., Groba, F., 2013. Chinese Renewable Energy Technology Exports: The Role of Policy, Innovation and Markets. DIW Berlin Discussion Paper No. 1263, available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2205645.
Choi, C., Park, Y., 2009. Monitoring the organic structure of technology based on the patent development paths. Technological Forecasting and Social Change 76, 754–768.
Clark, K.B., 1985. The interaction of design hierarchies and market concepts in technological evolution. Research Policy 14, 235–251.
Constant, E.W., 1980. The origins of the turbojet revolution. Johns Hopkins University Press, Baltimore, MD.
Davies, A., 1997. The life cycle of a complex product system. International Journal of Innovation Management 1, 229–256.
Davies, A., Hobday, M., 2005. The business of projects: managing innovation in complex products and systems. Cambridge University Press, Cambridge, UK.
Dechezleprêtre, A., Glachant, M., 2014. Does foreign environmental policy influence domestic innovation? Evidence from the wind industry. Environmental and Resource Economics 58, 391–413.
145
Dosi, G., 1982. Technological paradigms and technological trajectories: A suggested interpretation of the determinants and directions of technical change. Research Policy 11, 147–162.
Dosi, G., 1988. Sources, procedures, and microeconomic effects of innovation. Journal of Economic Literature 26, 1120–1171.
Dosi, G., Nelson, R.R., 2013. The Evolution of Technologies: An Assessment of the State-of-the-Art. Eurasian Business Review 3, 3–46.
Epicoco, M., 2013. Knowledge patterns and sources of leadership: Mapping the semiconductor miniaturization trajectory. Research Policy 42, 180–195.
Epicoco, M., Oltra, V., Saint Jean, M., 2014. Knowledge dynamics and sources of eco-innovation: Mapping the Green Chemistry community. Technological Forecasting and Social Change 81, 388–402.
Fleming, L., Sorenson, O., 2001. Technology as a complex adaptive system: evidence from patent data. Research Policy 30, 1019–1039.
Fontana, R., Nuvolari, A., Verspagen, B., 2009. Mapping technological trajectories as patent citation networks. An application to data communication standards. Economics of Innovation and New Technology 18, 311– 336.
Fraunhofer ISE, 2012. Photovoltaics Report. Fraunhofer Institute for Solar Energy Systems ISE, Freiburg, Germany.
Funk, J., 2009. Components, systems and discontinuities: The case of magnetic recording and playback equipment. Research Policy 38, 1192–1202.
Gallagher, K., Grubler, A., Kuhl, L., Nemet, G., Wilson, C., 2012. The Energy Technology Innovation System. Annual Review of Environment and Resources 37, 137–162.
Gallagher, K.S., 2014. The Globalization of Clean Energy Technology. MIT Press, Cambridge, MA.
Garrad, A., 2012. The lessons learned from the development of the wind energy industry that might be applied to marine industry renewables. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences 370, 451–71.
Garud, R., Karnøe, P., 2003. Bricolage versus breakthrough: distributed and embedded agency in technology entrepreneurship. Research Policy 32, 277–300.
Gipe, P., 1995. Wind energy comes of age. John Wiley & Sons Inc, New York, NY.
Gort, M., Klepper, S., 1982. Time paths in the diffusion of product innovations. Economic Journal 92, 630–653.
146 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
Griliches, Z., 1990. Patent statistics as economic indicators: a survey. Journal of Economic Literature 28, 1661– 1707.
Grubb, M., 2004. Technology Innovation and Climate Change Policy: an overview of issues and options. Keio Economic Studies 41, 103.
Grubler, A., 2012. Energy transitions research: Insights and cautionary tales. Energy Policy 50, 1–9.
Grubler, A., Wilson, C., 2014. Energy Technology Innovation: Learning from Historical Successes and Failures. Cambridge Univ Pr, Cambridge, UK and New York, NY.
Harhoff, D., Scherer, F.M., Vopel, K., 2003. Citations, family size, opposition and the value of patent rights. Research Policy 32, 1343–1363.
Hatch, N.W., Mowery, D.C., 1998. Process innovation and learning by doing in semiconductor manufacturing. Management Science 44, 1461–1477.
Hayami, Y., Ruttan, V., 1973. Professor Rosenberg and the Direction of Technological Change: A Comment. Economic Development and Cultural Change 21, 352–355.
Henderson, R., 1995. Of life cycles real and imaginary: The unexpectedly long old age of optical lithography. Research Policy 24, 631–643.
Ho, J.C., Saw, E.-C., Lu, L.Y.Y., Liu, J.S., 2014. Technological barriers and research trends in fuel cell technologies: A citation network analysis. Technological Forecasting and Social Change 82, 66–79.
Hobday, M., 1998. Product complexity, innovation and industrial organisation. Research Policy 26, 689–710.
Hoffmann, W., Peitruszko, S., Viaud, M., 2004. Towards an Effective European Industrial Policy for PV Solar Electricity. 19th PVSEC Paris 15–19.
Hoppmann, J., Peters, M., Schneider, M., Hoffmann, V.H., 2013. The two faces of market support—How deployment policies affect technological exploration and exploitation in the solar photovoltaic industry. Research Policy 42, 989–1003.
Huberty, M., Zachmann, G., 2011. Green Exports and the Global Product Space: Prospects for EU Industrial Policy Working Papers. Bruegel.
Huenteler, J., Niebuhr, C., Schmidt, T.S., 2014a. The effect of local and global learning on the cost of renewable energy in developing countries. Journal of Cleaner Production (in press).
Huenteler, J., Ossenbrink, J., Schmidt, T.S., Hoffmann, V.H., 2014b. How a Product’s Design Hierarchy Shapes the Evolution of Technological Knowledge – Evidence from Patent-Citation Networks in Wind Power. Working Paper, available at SSRN: http://ssrn.com/abstract=2534072.
147
Hummon, N.P., Doreian, P., 1989. Connectivity in a citation network: The development of DNA theory. Social Networks 11, 39–63.
ICTSD, 2010. Technology Mapping of the Renewable Energy, Buildings, and Transport Sectors: Policy Drivers and International Trade Aspects. International Centre for Trade and Sustainable Development (ICTSD), London, UK.
IEA, 2012. Trends in Photovoltaic Power Systems Programme - Survey Report of Selected IEA Countries between 1992 and 2011. International Energy Agency (IEA) - Photovoltaic Power Systems Program (PVPS), Paris, France.
IEA, 2013. World Energy Outlook 2013. International Energy Agency (IEA) / Organisation for Economic Cooperation and Development (OECD), Paris, France.
IEA, 2014. World Energy Outlook 2014 World Energy Outlook. International Energy Agency (IEA), Paris, France.
Jager-Waldau, A., 2004. Status of thin film solar cells in research, production and the market. Solar Energy 77, 667–678.
Kamp, L., 2004. Notions on learning applied to wind turbine development in the Netherlands and Denmark. Energy Policy 32, 1625–1637.
Karnøe, P., 1993. Approaches to innovation in modern wind energy technology: technology policies, science, engineers and craft traditions. Center for Economic Policy Research - Publication 334, Stanford, CA.
Lee, J., Berente, N., 2013. The era of incremental change in the technology innovation life cycle: An analysis of the automotive emission control industry. Research Policy 42, 1469–1481.
Lewis, J., 2007. A Comparison of Wind Power Industry Development Strategies in Spain, India and China. Studies in Comparative International Development 42, 208–232.
Lewis, J., 2011. Building a national wind turbine industry: experiences from China, India and South Korea. International Journal of Technology and Globalisation 5, 281–305.
Lewis, J.I., Wiser, R.H., 2007. Fostering a renewable energy technology industry: An international comparison of wind industry policy support mechanisms. Energy Policy 35, 1844–1857.
Liu, J., Goldstein, D., 2012. Understanding China’s renewable energy technology exports. Energy Policy 52, 417–428.
Magnusson, T., Tell, F., Watson, J., 2005. From CoPS to mass production? Capabilities and innovation in power generation equipment manufacturing. Industrial and Corporate Change 14, 1–26.
148 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
Markard, J., 2011. Transformation of infrastructures: sector characteristics and implications for fundamental change. Journal of Infrastructure Systems 17, 107.
Marsili, O., 2001. The anatomy and evolution of industries: Technological change and industrial dynamics New horizons in the economics of innovation. Edward Elgar, Cheltenham UK and , Northampton MA.
Martinelli, A., 2012. An emerging paradigm or just another trajectory? Understanding the nature of technological changes using engineering heuristics in the telecommunications switching industry. Research Policy 41, 414–429.
Martinelli, A., Nomaler, Ö., 2014. Measuring knowledge persistence: a genetic approach to patent citation networks. Journal of Evolutionary Economics 24, 623–652.
Menanteau, P., 2000. Learning from Variety and Competition Between Technological Options for Generating Photovoltaic Electricity. Technological Forecasting and Social Change 63, 63–80.
Menzel, M.-P., Kammer, J., 2011. Pre-entry Experiences, Technological Designs, and Spatial Restructuring in the Global Wind Turbine Industry. DIME Final Conference 6-8 April 2011, Maastricht, The Netherlands.
Miller, R., Hobday, M., Leroux-Demers, T., Olleros, X., 1995. Innovation in Complex Systems Industries: the Case of Flight Simulation. Industrial and Corporate Change 4, 363–400.
Mina, A., Ramlogan, R., Tampubolon, G., Metcalfe, J.S., 2007. Mapping evolutionary trajectories: Applications to the growth and transformation of medical knowledge. Research Policy 36, 789–806.
Moore, L., Post, H., 2008. Five years of operating experience at a large, utility-scale photovoltaic generating plant. Progress in Photovoltaics: Research 249–259.
Mowery, D.C., Rosenberg, N., 1979. The influence of market demand upon innovation: a critical review of some recent empirical studies. Research Policy 8, 102–153.
Murmann, J.P., Frenken, K., 2006. Toward a systematic framework for research on dominant designs, technological innovations, and industrial change. Research Policy 35, 925–952.
Murmann, J.P., Tushman, M.L., 2002. Dominant Designs, Technology Cycles, and Organizational Outcomes, in: Garud, R., Kumaraswamy, A., Langlois, R.N. (Eds), Managing in the Modular Age: Architectures, Networks, and Orgaetworks, and Organizations. Blackwell, Malden, MA.
Neij, L., 1997. Use of experience curves to analyse the prospects for diffusion and adoption of renewable energy technology. Energy Policy 23, 1099–1107.
Nemet, G.F., 2006. Beyond the learning curve: factors influencing cost reductions in photovoltaics. Energy Policy 34, 3218–3232.
149
Nemet, G.F., 2009. Demand-pull, technology-push, and government-led incentives for non-incremental technical change. Research Policy 38, 700–709.
Nightingale, P., 2000. The product–process–organisation relationship in complex development projects. Research Policy 29, 913–930.
De Nooy, W., Mrvar, A., Batagelj, V., 2011. Exploratory Social Network Analysis with Pajek - Revised and Expanded. Cambridge University Press, Cambridge, UK.
Norberg-Bohm, V., 2000. Creating Incentives for Environmentally Enhancing Technological Change: Lessons From 30 Years of US Energy Technology Policy. Technological Forecasting and Social Change 65, 125– 148.
Pavitt, K., 1984. Sectoral patterns of technical change: towards a taxonomy and a theory. Research Policy 13, 343–373.
PCAST, 1999. Report to the President on Powerful Partnerships: The Federal Role in International Cooperation on Energy Innovation. The President’s Committee of Advisors on Science and Technology (PCAST), Washington, DC.
Peltoniemi, M., 2011. Reviewing Industry Life-cycle Theory: Avenues for Future Research. International Journal of Management Reviews 13, 349–375.
Pernick, R.O.N., Wilder, C., Winnie, T., 2013. Clean Energy Trends 2013. Clean Edge, Inc., Portland, OR.
Peters, M., 2011. The Role of Technology Policy in Fostering Technical Change – Lessons from the Cases of Solar and Wind Power. Ph.D. Dissertation, ETH Zürich.
Peters, M., Schneider, M., Griesshaber, T., Hoffmann, V.H., 2012. The impact of technology-push and demand-pull policies on technical change – Does the locus of policies matter? Research Policy 41, 1296– 1308.
Pielke, R., Wigley, T., Green, C., 2008. Dangerous assumptions. Nature 452, 531–532.
Pizer, W., Popp, D., 2008. Endogenizing technological change: Matching empirical evidence to modeling needs. Energy Economics 30, 2754–2770.
Polanyi, M., 1962. The Republic of Science: Its Political and Economic Theory. Minerva 1, 54–73.
Prabhakaran, T., Lathabai, H.H., Changat, M., 2014. Detection of paradigm shifts and emerging fields using scientific network: A case study of Information Technology for Engineering. Technological Forecasting and Social Change (in press).
150 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
Prencipe, A., 2000. Breadth and depth of technological capabilities in CoPS: the case of the aircraft engine control system. Research Policy 29, 895–911.
Rosenberg, N., 1969. The direction of technological change: inducement mechanisms and focusing devices. Economic Development and Cultural Change 18, 1–24.
Rosenberg, N., 1982. Inside the black box: technology and economics. Cambridge University Press, Cambridge, UK.
Rosenkopf, L., Nerkar, A., 1999. On the Complexity of Technological Evolution - Exploring Coevolution Within and Across Hierarchical Levels in Optical Disc Technology, in: Baum, J.A.C., McKelvey, B. (Eds), Variations in Organization Science: In Honor of D. T. Campbell. SAGE Publications, Inc, New York, NY.
Rosenkopf, L., Nerkar, A., 2001. Beyond local search: boundary spanning, exploration, and impact in the optical disk industry. Strategic Management Journal 22, 287–306.
Routley, M., Phaal, R., Probert, D., 2013. Exploring industry dynamics and interactions. Technological Forecasting and Social Change 80, 1147–1161.
Suarez, F., Utterback, J., 1993. Patterns of industrial evolution, dominant designs, and firms’ survival, in: Burgelman, R., Rosenbloom, R. (Eds), Research on Technological Innovation, Management and Policy. JAI Press, Inc, Greenwich, CT, pp. 79–92.
Tang, T., Popp, D., 2013. The Learning Process and Technological Change in Wind Power: Evidence from China’s CDM Wind Projects. 0–47.
The White House, 2009. The Recovery Act. Washington, DC.
Trancik, J., 2006. Scale and innovation in the energy sector: a focus on photovoltaics and nuclear fission. Environmental Research Letters 1, 1–7.
Twidell, J., 2009. Wind Energy-The Facts. Wind Engineering 33, 305–307.
Utterback, J.M., Abernathy, W.J., 1975. A dynamic model of process and product innovation. Omega 3, 639– 656.
Utterback, J.M., Suárez, F.F., 1993. Innovation, competition, and industry structure. Research Policy 22, 1–21.
Vernon, R., 1966. International investment and international trade in the product cycle. The Quarterly Journal of Economics 80, 190–207.
Verspagen, B., 2007. Mapping technological trajectories as patent citation networks: A study on the history of fuel cell research. Advances in Complex Systems 10, 93–115.
151
Vincenti, W., 1990. What engineers know and how they know it: Analytical studies from aeronautical history. Johns Hopkins University Press, Baltimore, MD.
Wiesenthal, T., Leduc, G., Haegeman, K., Schwarz, H.-G., 2011. Bottom-up estimation of industrial and public R&D investment by technology in support of policy-making: The case of selected low-carbon energy technologies. Research Policy 41, 116–131.
Wilson, C., 2012. Up-scaling, formative phases, and learning in the historical diffusion of energy technologies. Energy Policy 50, 81–94.
Winskel, M., Markusson, N., Jeffrey, H., 2014. Learning pathways for energy supply technologies: Bridging between innovation studies and learning rates. Technological Forecasting & Social Change 81, 96–114.
Van Zeebroeck, N., 2011. The puzzle of patent value indicators. Economics of Innovation and New Technology 20, 33–62.
152 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
Appendix
Table A1: Main engineering tasks in solar PV product and process development (areas of PV-specific knowledge are shaded in grey)
System element Product design Production process
Process, equipment and plant design for production of cell materials Design of cell materials and Process, equipment and plant design for production of solar cell; surface Solar cell arrangement treatment; contact printing Design of electrical contact patterns Design of optical and electrical testing equipment
Design of module circuitry Process, equipment and plant design for cell interconnection, encapsulation, Module Design of encapsulation materials, aluminum frame and glass processing back cover and frame Design of optical and electrical testing equipment
Design of load carrying structures and control system Metalworking and assembly Mounting system Transport-, installation-, and Electronics manufacturing and assembly O&M-friendly design
Design and dimensioning of control Grid connection Electronics manufacturing and assembly and power electronics
Table A2: Main engineering tasks in product and process development wind power (areas of wind-specific knowledge are shaded in grey)
System element Product design Production process
Processing of composites and core materials Development of structural materials and coating Design of specialized molds Aerodynamic and structural design Design of non-destructive testing equipment Rotor Choice of rotor control and procedures Design and integration of electric motors, gears, hydraulics, control Metalworking, electrical manufacturing and systems and power sources assembly
Design of mechanical drive-train architecture Dimensioning and material selection for hub, bearings, shafts, brakes, Metalworking and assembly gearbox, lubrication, joints and couplings Electrical equipment manufacturing and Power train Choice of generator topology assembly Design and dimensioning of generator, power electronics, cooling and Electronics manufacturing and assembly control systems
Design of load transfer, noise insulation and thermal management Composite processing (thermal and chemical Aesthetic and aerodynamic design process engineering) Mounting & Transport-, installation-, and O&M-friendly design Metalworking encapsulation Dimensioning of tower and foundation for static and dynamic load Steel processing transfer Concrete production
Design of wind-farm circuitry, voltage transfer, electrical insulation Electrical equipment manufacturing and Choice and design of storage technology Grid connection assembly Design of control strategy and software Electronics manufacturing and assembly Design and integration of control system elements
153
Table A3: Patents along critical path of solar PV citation network 1963-2009
Priority Focus of Application Focus of invention Assignee Assignee type patent invention
US 15-Apr-74 Cell concept (polycrystalline silicon) Cell (product) E. Crisman Individual 3,978,333
US 28-Jul-75 Cell concept (amorphous silicon) Cell (product) RCA Cell manufacturer 4,064,521
US 28-Mar-77 Non-reflecting surface layers for solar cell Cell (product) RCA Cell manufacturer 4,126,150
US 24-Apr-78 Cell concept (amorphous silicon) Cell (product) RCA Cell manufacturer 4,162,505
US Cell concept (tandem junction amorphous 19-Apr-79 Cell (product) RCA Cell manufacturer 4,272,641 silicon)
US Module Energy Conversion Devices 11-Feb-82 Procedure to connect cells in module Cell manufacturer 4,419,530 (process) Inc.
US Module Energy Conversion Devices 9-Nov-82 Cell interconnection in module Cell manufacturer 4,443,652 (product) Inc.
US Energy Conversion Devices 7-Nov-83 Substrate sheet for thin-film module Cell (product) Cell manufacturer 4,514,583 Inc.
US Cell manufacturer 30-Oct-85 Substrate sheet for thin-film module Cell (product) Astrosystems Inc. 4,677,250
US Production process for polycrystalline thin- 26-Jan-87 Cell (process) Canon Cell manufacturer 5,087,296 film cell
US Production process for crystalline thin-film 24-Aug-87 Cell (process) Canon Cell manufacturer 5,130,103 cell
US Production process for crystalline thin-film 16-Jun-89 Cell (process) Canon Cell manufacturer 5,094,697 cell
US Production process for polycrystalline thin- 26-Dec-90 Cell (process) Canon Cell manufacturer 5,403,771 film cell
US Production process for crystalline thin-film 10-Mar-94 Cell (process) Canon Cell manufacturer 5,856,229 cell
US Production process for silicon-on-insulator 10-Mar-94 Cell (process) Canon Cell manufacturer 5,854,123 cell
US Production process for crystalline thin-film 2-Feb-95 Cell (process) Sony Cell manufacturer 6,326,280 cell
US Production process for silicon-on-insulator 28-Feb-96 Cell (process) Canon Cell manufacturer 6,294,478 cell
US Production process for silicon-on-insulator 15-Nov-96 Cell (process) Canon Cell manufacturer 6,054,363 cell
US Production process for silicon-on-insulator 26-Mar-97 Cell (process) Canon Cell manufacturer 6,221,738 cell
US Production process for silicon-on-insulator Production equipment 12-May-97 Cell (process) Silicon Genesis Corp. 6,582,999 cell provider
US Production process for silicon-on-insulator 15-May-98 Cell (process) Canon Cell manufacturer 6,613,678 cell
US 8-Jun-99 Production process for microcrystalline cell Cell (process) Canon Cell manufacturer 6,664,169
US 16-Aug-00 Production process for silicon-germanium- Cell (process) Massachusetts Institute of Public sector
154 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
6,573,126 on-insulator based cell Technology
US Production process for a substrate for thin- 27-Nov-00 Cell (process) Soitec Technologies Cell manufacturer 6,794,276 film solar cell
US Production process for germanium California Institute of 17-Apr-01 Cell (process) Public sector 7,019,339 heterostructure cell Technology
US Production process for silicon heterostructure California Institute of 17-Apr-01 Cell (process) Public sector 7,341,927 cell Technology
US 21-Oct-04 Multi-junction cell concept Cell (product) Aonex Technologies Materials supplier 7,846,759
US Production equipment 27-Jul-05 Production process for thin-film cell Cell (process) Silicon Genesis Corp. 7,911,016 provider
US Production equipment 5-Apr-06 Production process for thin-film cell Cell (process) Silicon Genesis Corp. 7,759,220 provider
US Production process for microcrystalline silicon Production equipment 23-Jun-06 Cell (process) Applied Materials 7,655,542 cell provider
US Production process for thin-film multi- Production equipment 18-Jan-07 Cell (process) Applied Materials 8,203,071 junction cell provider
US Production equipment 10-Jul-07 Production process for thin-film cell Cell (process) Applied Materials 7,875,486 provider
US Production equipment 31-Aug-07 Method of forming contacts on thin-film cell Cell (process) Applied Materials 7,908,743 provider
US 5-Mar-08 Production process for thin-film cell Cell (process) Global Solar Energy Cell manufacturer 8,062,922
US 24-Jul-09 Production process for thin-film cell Cell (process) Solopower Cell manufacturer 8,318,530
155
Table A4: Patents along critical path of wind-patent citation network 1963-2009
Priority Application Focus of invention Focus of invention Assignee Assignee type patent
Blade with integrated over-speeding control Svenning Konsult Engineering SE 005,407 12-May-75 Rotor (product) mechanism AB consultancy
DE Rotor-hub arrangement with teetering hub and two U. Huetter 4-Dec-76 Rotor (product) Individual 2,655,026 blades (Indiv.)
US Control system for two-bladed rotor with adjustable Turbine 8-Jun-78 Rotor (product) MAN 4,297,076 tips manufacturer
US Three-bladed turbine with hydraulic pitch C E Kenney 31-Jul-78 Rotor (product) Individual 4,274,807 mechanism (Indiv.)
US Two-bladed downwind turbine with teetering hub Carter Wind Turbine 10-May-79 Rotor (product) 4,366,387 and aerodynamic pitch mechanism Power manufacturer
US Rotor with teetered hub and mechanical pitch North Wind Turbine 24-Feb-82 Rotor (product) 4,435,646 control system Power manufacturer
US Two-blade turbine with novel drag brake and Turbine 29-Sep-83 Rotor (product) Boeing 4,565,929 control system manufacturer
US Torque control system for variable-speed power United Turbine 18-Nov-85 Power train (product) 4,703,189 train Technologies manufacturer
US United Turbine 28-Apr-86 Operation strategy for variable-speed power train Power train (product) 4,700,081 Technologies manufacturer
US Variable-speed power train architecture and power Turbine 1-Feb-91 Power train (product) US WindPower 5,083,039 control manufacturer
US Turbine 19-Sep-91 Speed control system for variable-speed power train Power train (product) US WindPower 5,155,375 manufacturer
US 6-Feb-95 Power train control for variable wind conditions Power train (product) U.S. EPA Public sector 5,652,485
US Variable-speed power train architecture and power Zond Energy Turbine 8-Aug-97 Power train (product) 6,137,187 control Systems manufacturer
US Variable-speed power train adapted to smoothen Vestas Wind Turbine 23-May-00 Power train (product) 6,566,764 power output Systems manufacturer
US Inverter control system for grid-friendly power Grid connection 10-Jul-01 ABB Generator supplier 6,670,721 output (product)
DE Collective control method for turbines in a wind Grid connection Turbine 28-Sep-01 Enercon 1,048,225 farm (product) manufacturer
US 8-Apr-03 Variable-speed power train architecture Power train (product) Alstom Generator supplier 7,190,085
US Clipper Turbine 7-May-03 Variable-speed power train architecture Power train (product) 7,042,110 Windpower manufacturer
US Generator control optimizing response to grid Grid connection Turbine 8-Jan-04 Hitachi 7,205,676 failure (product) manufacturer
Mounting & Mitsubishi Turbine JP 055,515 27-Feb-04 System to control nacelle vibrations encapsulation (product) HeavyInd. manufacturer
US 30-Sep-04 System to control turbine vibrations Mounting & General Electric Turbine
156 Technology Life-Cycles in the Energy Sector – Technological Characteristics and the Role of Deployment for Innovation
7,309,930 encapsulation (product) manufacturer
US Power train control routine based on upstream wind Turbine 30-Sep-05 Power train (product) General Electric 7,342,323 measurements manufacturer
US Mounting & Fuji Heavy Turbine 1-Feb-06 Control routine to suppress tower vibrations 7,400,055 encapsulation (product) Industries manufacturer
US Grid connection Turbine 14-Sep-06 Control routine to respond to grid faults Vestas 7,851,934 (product) manufacturer
US Grid connection Turbine 14-Sep-06 Control routine to respond to grid faults Vestas 7,911,072 (product) manufacturer
US Control routine to respond to grid-side load Grid connection Turbine 22-Feb-08 Nordex 7,714,458 shedding (product) manufacturer
US Control system for wind farm with redundant Grid connection Turbine 16-Jun-08 Nordex 7,949,434 control unit (product) manufacturer
157
a) Solar PV 1970 1980 1990 2000 2005 2009
PV system architecture
Product Cell Process
Product Module Process
Mounting system Product Process
Grid connection Product Process
Patent relating to product innovation (size ~ node weight) Citation ≤ 5 year lag (width ~ link weight) Legend: Patent relating to proces innovation Citation > 5 year lag
b) Wind power 1975 1980 1990 2000 2005 2009
Wind turbine architecture
Product Rotor Process
Product Power train Process
Grid connection Product Process
Mounting & Product encapsulation Process
Figure A 1: Patents in 80%-weight network (full networks D in Table 4) ordered by time of patent filing and their focus in the technological system; linkages indicate citations.
158
The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries
Joern Huenteler1,2*, Christian Niebuhr1,3, Tobias S. Schmidt1,4
1Department of Management, Technology and Economics, ETH Zurich, Switzerland 2Belfer Center for Science and International Affairs, John F. Kennedy School of Government, Harvard University, USA 3School of Business and Economics, RWTH Aachen University, Germany 4Precourt Energy Efficiency Center, Stanford University, USA *corresponding author: [email protected]; +41 44 632 97 39; +41 44 632 05 41
Forthcoming in the Journal of Cleaner Production (DOI: 10.1016/j.jclepro.2014.06.056). Previous versions of this paper have been presented at the UNFCCC COP 19 in Warsaw, Poland, the workshop on “Governance Architecture towards Low-Carbon Society: Technology and Actor Configuration” at the United Nations University in Yokohama, Japan, in October 2013, and the “Sustainability Innovation Seminar” at the University of Tokyo, Japan, in February 2013.
Abstract
High upfront costs are a critical barrier for investments in clean infrastructure technologies in developing countries. This paper uses a case study of Thailand’s electricity sector to create realistic estimates for the relative contributions of local and global technological learning to reducing this cost in the future and discusses implications of such learnings for international climate policy. For six renewable electricity technologies, we derive estimates for the share of locally and globally sourced goods and services, and analyze the effects of local and global learning during the implementation of Thailand’s renewable energy targets for 2021. Our results suggest that, in aggregate, the largest potential for cost reduction lies in local learning. This finding lends quantitative support to the argument that the conditions enabling local learning, such as a skilled workforce, a stable regulatory framework, and the establishment of sustainable business models, have a more significant impact on cost of renewable energy in developing countries than global technology learning curves. The recent shift of international support under the United Nations Framework Convention on Climate Change towards country-specific technology support is therefore promising. However, our results also show that the relative importance of local and global learning differs significantly between technologies, and is determined by technology and country characteristics. This suggests that international support need to consider both the global perspective and local context and framework conditions in order to reap the full benefits of technological learning across the wide range of clean technologies.
Keywords: Climate policy, Technology transfer, Technological capabilities, Technological learning, Thailand, Renewable energy
160 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries
Highlights:
Development of a techno-economic model of Thailand’s Alternative Energy Development Plan 2012-2021. Analysis of the impacts of local and global learning effects on mitigation cost. Demonstration that the importance of global and local learning varies between clean technologies. Finding that local learning is significant for wind, PV, biogas and micro hydro, whereas global learning is important for PV and solar thermal. Discussion of the future role of the international support for clean technologies.
161
1. Introduction
The global climate policy regime needs to significantly accelerate the diffusion of clean technologies to avoid dangerous impacts from climate change (UNFCCC, 2012). In addition to actions taken by the developed world, developing countries are expected to assume greater responsibility by implementing domestic policies that contribute to both domestic economic development and climate change mitigation (Kanie et al., 2010). Indeed many developing countries are already implementing domestic climate legislation, despite the gridlock in international negotiations (Nachmany et al., 2014; REN21, 2013; Townshend et al., 2013). However, high upfront costs remain a critical barrier for large-scale investments in clean technologies, especially in developing countries (IPCC, 2012; Schmidt, 2014). How to accelerate the development and transfer of clean technologies is, therefore, emerging as a central issue in the international climate policy negotiations (Ockwell and Mallett, 2012; Pueyo et al., 2012).
Experience in the industrialized world has shown that cost reductions and performance improvements of new technologies are often closely linked to policies aimed at increased production and deployment (Jänicke, 2012), driven by mechanisms collectively referred to as technological learning (Junginger et al., 2010). If successful, the increasing number of mitigation actions taken now by developing countries holds the promise to stimulate innovations and future cost reductions there as well. But technological learning encapsulates a diverse array of purposeful processes that some countries, sectors and organizations manage better than others (Bell and Figueiredo, 2012; van Hoof, 2014). Besides creating financial incentives for investment, one of the key challenges for international climate policy is therefore to actively promote technological capabilities in developing countries and to enable countries to reap the full learning benefits from mitigation investments they make and attract (Benioff et al., 2010; Bhasin, 2013; de Coninck et al., 2008; Ockwell and Mallett, 2012).
Technological learning in developing countries, especially outside the largest emerging economies, follows distinct dynamics (Pueyo et al., 2011). The industries producing clean technologies are increasingly globalized (Gallagher, 2014; Lewis, 2012; Nahm and Steinfeld, 2014). Therefore, in a typical investment project, local firms in developing countries provide only part of the products and services. Learning in this share of the industry value chain is local in nature and driven by local market developments and policies – we will refer to it as local technological learning (Morrison et al., 2008; Mytelka, 2000). However, because a substantial share of components is typically sourced from abroad, the economics of local investments are also impacted by technological learning processes in other
162 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries countries. For example, technological progress by Chinese solar cell producers improves the economics of solar investments around the world. This form of learning is driven more by global markets than by policies in individual countries (Peters et al., 2012). Future investment conditions for clean technologies in developing countries thus depend on a combination of global and local learning processes, which, in turn, depend on domestic and international regulatory, institutional and industrial contexts. Better understanding of the relative importance of the two can improve both domestic and international policy decisions.
Using a quantitative case study, this paper estimates the effect of local and global technological learning on the cost reductions of renewable electricity generation in Thailand. We employ a techno- economic model of the country’s electricity sector to project the cost of implementing the country’s renewable energy targets for 2021 (Kamolpanus, 2013). We derive estimates for the share of locally and globally sourced goods and services for six renewable electricity technologies and analyze, in different scenarios, the impact of local and global learning effects on the investment cost. Based on the results, we explore implications for the design of international low carbon technology support mechanisms.
The paper makes three main contributions. First, our case study informs the academic debate as well as international negotiations on the post-Kyoto climate policy regime of the United Nations Framework Convention on Climate Change (UNFCCC). In its support for technology development and transfer, the international climate policy regime has recently shifted its attention toward national policies and local technological learning. The analysis presented in this paper enhances the understanding of the merits of this shift, and informs the design and functional specification of the new international technology support mechanisms. Our quantitative approach and the focus on mitigation cost complements existing conceptual and qualitative work on the topic (Benioff et al., 2010; Bhasin, 2013; de Coninck et al., 2008; Ockwell and Mallett, 2012). Furthermore, it contributes to the growing body of literature on the economics of clean energy technology investments in developing countries (e.g., IRENA, 2012a; Schmidt et al., 2012). Finally, our paper is among the first to investigate the impact of local and global learning separately for a specific developing country case.
The next section will introduce the key theoretical constructs used in the analysis (Section 2). Section 3 introduces the case, before section 4 presents the model, the data sources, and the methodology. The results of the case study are presented in Section 0, and their policy implications discussed in Section 0.
163
2. Local and Global Technological Learning
2.1. Technological Learning in Developing Countries
Technological learning is understood here broadly as the accumulation of technological knowledge and experience, often also referred to as technological capabilities, in individuals and organizations (Bell and Figueiredo, 2012). Research on innovation processes has shown that the technological capabilities held by firms comprises not only information codified in capital goods or documents (patents, manuals, etc.), but also includes the tacit knowledge embodied in individual skills and firm routines (Dosi, 1988; Senker, 1995). These elements of knowledge are costly to transfer and therefore highly organization- specific (von Hippel, 1994). This means that removing trade barriers and providing developing countries with intellectual property rights (IPR) and resources for technology imports is not sufficient to enable countries to catch up to the technological frontier (Bell and Pavitt, 1996; Ockwell et al., 2010). Rather, catching up requires building local technological capabilities through the cumulative, costly and time-consuming process of technological learning (Bell, 2010).
Technological capabilities and learning are increasingly being recognized as significant drivers of low carbon development (Byrne et al., 2011; Lema and Lema, 2013; Phillips et al., 2013). The international climate negotiations, too, are taking notice (Ockwell and Mallett, 2012). Improved technological capabilities hold the promise of removing barriers to the diffusion of clean technologies, thereby facilitating further emission reductions in the future (Sandén and Azar, 2005). Besides its effect on mitigation cost, the local build-up of technological capabilities is crucial for local industrial capacity, poverty reduction and economic growth. For many developing countries, investing in climate change mitigation is, for now, only desirable if the government can create opportunities for the local private sector to participate in the value chain of mitigation investments. However, in order to participate in the development and manufacturing of clean technologies, local firms in developing countries need to create the capacity to continuously absorb, adapt and improve new technologies (Bell and Pavitt, 1996).
Climate models increasingly incorporate learning as an endogenous process driven by mitigation investments (Kahouli-Brahmi, 2008; van der Zwaan et al., 2002), but technological learning is not an automatic by-product of investments (Bell and Figueiredo, 2012). Rather, in the analysis of the development of mitigation policies and estimation of future mitigation cost, it is better understood as an opportunity that can be only adequately seized when both governments and firms create the
164 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries necessary conditions. Organizations need to pursue conscious efforts to create the ability, in the form of a skilled workforce and organizational processes, to absorb the new knowledge and experience that they generate (Cohen and Levinthal, 1989). Furthermore, organizations innovate and learn through their interaction with users, suppliers, competitors, universities or regulators in systems of innovation (Fagerberg et al., 2007; Lundvall et al., 2009). The existence of formal and informal networks, as well as public funding for science and technology, are therefore critical drivers of technological learning. And, last but not least, learned capabilities degenerate rapidly if organizations have a rapid workforce turnover, face an instable regulatory framework, or pursue unsustainable business models.
2.2. Local and Global Learning Effects in Value Chains
Most clean technologies are technological systems consisting of hundreds, or even thousands, of materials, components, and intermediate goods. Furthermore, mitigation investments involve numerous legal, financial, and regulatory services. The collective of technology suppliers and service providers that deliver the materials, components, products and services to deploy technologies we call the technologies’ industry value chain.
Modern industry value chains are disintegrated and geographically distributed production and service networks. As markets for clean technologies have grown, their supplying industries have also globalized in recent years (e.g., Gallagher, 2014; Lewis, 2012; Nahm and Steinfeld, 2014). Globally traded components and products are often those that can be transported at relatively low cost and have standardized interfaces. (On the extreme end of this spectrum are commodities.) Globally traded services often require highly specialized technological expertise that only very few firms possess. In the wind turbine industry, for example, gearboxes, hubs, generators and bearings are components for which the know-how necessary for design and manufacturing is concentrated in only a few key firms globally. Further, the consulting services needed to make a complex product bankable, or a difficult geography accessible, are often provided by experienced, globally operating firms.
For technological learning in globally traded goods and services, global market conditions matter more than where their products are finally deployed. If uncertainty about the product’s performance is very large, as in the case of carbon capture and storage technology, any demonstration will add to the global knowledge pool (de Coninck et al., 2009). In the case of smaller components, materials, or intermediate goods, producers seldom even interact with end-users, if they know them at all. In the case of services, the global applicability of their experience is the key reason why globally operating producers are selected in the first place. The accumulation of capabilities in firms and industries is
165
therefore more dependent on global, aggregate market trends than on country-specific context factors. We define learning in these goods and services as global technological learning, because, in a simplified learning model or experience curve, it would best be predicted by the size of the global market.
But value chains are not entirely globalized. To stimulate local private sector participation, many climate-related laws in developing countries contain some form of provision to create a certain level of domestic content. This gives local firms economic advantages over global suppliers. But even without legislative requirements, often local firms provide many steps. This may include large and heavy components that are costly to transport, or factors which are cheaper to make at home; but it is important to note that the drivers underlying these patterns are not affected only by input or transportation cost. The drivers for localization may include the expertise required to deal with idiosyncratic geography, context-specific or fast-changing regulations, or local infrastructure and climate conditions. In a wind power project, for example, towers, blades, and foundations are typically sourced from suppliers not far from the project site, and domestic firms often provide project development, installation, operation, and maintenance services. In developing countries, it is reasonable to assume that local firms are mostly active in their home markets. We will therefore assume that their learning is predominantly local, and refer to technological learning in this part of the value chain as local technological learning.
The geographical dispersion of value chains leads to cost and cost trends that differ significantly between components (e.g., Lindman and Söderholm, 2011). Cost trends and learning curves are global whenever global markets exists, while for the locally sourced components trends differ substantially between regions (e.g., Seel et al., 2014). For the latter, local economic, political, and regulatory conditions determine whether or not investments lead to the accumulation of technological capabilities, which in turn are essential to reduce local investment cost. To stimulate progress in this part of the value chain, domestic and international policymakers should focus on strengthening the domestic innovation system. For the global part of the value chain, however – which also affects domestic investment economics – national innovation systems are not very important. Here, policymakers need to work toward international knowledge sharing and standardization activities to strengthen the sectoral innovation system in order to advance low-carbon technologies (de Coninck et al., 2009). They should also strive for the global markets to remain open and try to minimize protectionism to reap the benefits of global technological learning (Lewis, 2014).
166 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries
3. The Case of Thailand’s Electricity Sector
3.1. Case Selection
This paper presents a quantitative case study of Thailand’s Alternative Energy Development Plan (DEDE, 2012) for the electricity sector in order to explore the relative importance of local and global learning in developing countries’ mitigation efforts. We chose a case study of the electricity sector because it lies at the heart of the climate change challenge as the single largest source of CO2 emissions among the primary sectors of the world economy (Bazilian et al., 2008). Indeed, the majority of national mitigation policies in developing countries target energy production and consumption in the industrial, energy supply, buildings, and transport sectors (van Tilburg et al., 2013). At the same time, the diversity of technologies in the electricity sector and their globally operating technology providers allows us to model both local and global learning processes.
We chose Thailand as case study for three reasons. First, the country has clear, broad and ambitious targets for renewable energy diffusion which allow us to study the impact of different learning conditions on the cost of an existing policy. Second, the country’s government publishes detailed data on energy production and consumption that allow us to model the electricity sector on a single-plant level. Third, the country faces economic and political challenges that make the framework conditions for its energy policy decisions representative of a large number of other middle-income countries. A country of 66.9 million with a GDP per capita at USD 5,210, Thailand has managed to provide its population with almost universal access to electricity (The World Bank, 2014). Like many other middle-income countries, it now faces the challenge of rapidly growing energy consumption, accompanied by growing carbon emissions, import dependency, national security concerns and local resistance to fossil and nuclear power plants. How to assist emerging economies in managing these challenges while simultaneously reducing carbon emissions will be one of the most important questions for international climate policy in the coming decades.
3.2. Trends and Challenges
Primary energy consumption in Thailand has almost tripled from 1990 to 2011, making it the second- largest energy consumer in the Association of Southeast Asian Nations (ASEAN), while subsequently its greenhouse gas (GHG) emissions grew by 177.5% (see Figure 1). The power sector is the largest carbon source, with a share in national emissions that grew from 33% in 1990 to 42% in 2011. By
167
2035, energy consumption and GHG emissions are expected to roughly double yet again (IEA, 2013a).
Thailand is already a net importer of oil, gas and coal, and is projected to become the most energy import-dependent country among the ASEAN by 2035, with imports estimated to increase to about 90% of consumed oil and gas (IEA, 2013a). Nakawiro et al (2008) estimate that gas and coal import costs will grow from 0.92% of the country’s GDP in 2011 to 2.19-2.69% in 2025, depending on the development of fuel prices in the region. The main domestic sources are not without challenges, too, in light of strong local opposition to nuclear power and new coal plants (Greacen and Bijoor, 2007; Pongsoi and Wongwises, 2013).44
Carbon emissions in Thailand
a) Historic emissions [in million tons CO2-eq] b) Projections, 2011=100
OtherIndustry Transport Power Generation GDP Power generation CO2-eq emissions from power generation1 250 250
200 200
150 150
100 100
50 50
0 0 1987 1990 1995 2000 2005 2010 2011 2015 2020 2025 2030
Figure 1: Thailand’s carbon emissions by sector (a) and future projections for emissions from the country’s power sector (b).
1 CO2 emissions from power sector do not contain imports. Data from EPPO (2013, 2012a).
As of 2011, the electricity sector is dominated by natural gas (67%), with lignite and hard coal providing together about 20% as well (Figure 2). Besides large hydropower (5%), renewable energy constitutes only a very small part of the electricity mix, mostly in the form of biomass (1.4%) (EPPO, 2012b). The remaining demand is covered by direct electricity imports (6.6%). Electricity generation reached 162 TWh in 2011 and is projected to increase by more than 4% annually (EPPO, 2012b, 2012c). Besides domestic capacity investments, the government plans to meet demand by increasing the share of direct electricity imports from neighboring Malaysia and Laos to 13% in 2030 (Figure 2).
44 The first nuclear plant was originally scheduled to come online in 2020, but was postponed to 2026 after the nuclear incidence in Fukushima, Japan, in 2011.
168 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries
Planned fuel mix of the electricity sector in Thailand 1987-2030 Electricity production by source, TWh
Imported electricity Renewables1 Hydro Fuel oil & diesel Coal Lignite Natural gas
350 Renewables Estimates Hydro 300 Imported 5% 1% Fuel oil & diesel 7% 250 1% Coal 8% 200
150 Lignite 12%
100 67% Natural gas 50
0 1990 2000 2010 2020 2030 2011 (total :162.3 TWh)
Figure 2: Development of the fuel mix in Thailand’s power sector from 1987 to 2030. 1The renewables wedge contains hydropower after 2011; the share of renewables after 2012 reflects relatively conservative projections of the Power Development Plan. Data from EPPO (2013, 2012a).
3.3. Targets and Support for Renewable Energy
In recent years, electricity sector planning initiatives have begun to consider renewable energy as a potential remedy for some of the problems the country faces. Thailand has no official renewable energy law at this point but several comprehensive long-term energy plans (Tongsopit and Greacen, 2013). The two most important are the Power Development Plan by the Energy Policy and Planning Office (EPPO, 2012c) and the Alternative Energy Development Plan (AEDP) by the Department of Alternative Energy Development and Efficiency (DEDE, 2012; Kamolpanus, 2013), both under the Ministry of Energy. The AEDP, updated in 2013, is aiming to increase the renewable energy in the power sector to 14 GW by 2021, or 24% of the total capacity (compare with Figure 2). As shown in Figure 3, the largest part of this capacity is projected to come from biomass (4.8 GW), followed by biogas (3.6 GW), solar power (3 GW), wind power (1.8 GW) and micro hydro (324 MW).45 The largest relative increase is targeted for biogas (17-fold) and wind energy (15-fold). It is notable that large hydro is not part of the AEDP. For simplification, we therefore use the term ‘renewable electricity’ in this paper to refer to non-large-hydro renewable electricity technologies.
45 The targets include another 400 MW of municipal waste incineration plants.
169
Targets for renewable energy development in Thailand in 2012-2021 Electric power generation capacity by source, MW Status end of 2012 Targets for 2021
+145% +1,761% 4,800 MW 13,924 MW +697% 3,600 3,000 +1,511% +400% 1,960 MW 1,800 +836% +218% 2,786 MW 324 193 377112 43 400 102
Biomass Biogas PV Wind MSW Micro Total renewables Hydro (excluding large hydro)
Figure 3: Targets for renewable electricity under Thailand‘s Alternative Energy Development Plan (DEDE, 2012). Data for 2012 are from DEDE (2013); updated targets for 2021 from Kamolpanus (2013).
In addition to public research, tax incentives, venture capital and investment grants, the primary government policy to induce renewable energy investments is currently a feed-in tariff premium scheme, referred to as ‘FIT adder’ (DEDE, 2012; Tongsopit and Greacen, 2013). The FIT adder program provides a purchase guarantee under which fixed premiums, which differ by technology, capacity and project location, are paid on top of a base tariff that is determined by the utility’s avoided cost. Originally implemented in 2007, the official objectives of the FIT adder policy included enhanced levels of renewable energy generation; private sector involvement; economic growth and rural development; diversification of the fuel mix; local pollution reduction and utilizing agricultural wastes; as well as local equipment manufacturing and thus reduced international equipment imports (Tongsopit and Greacen, 2013). In 2010, Thailand’s government announced plans to transform the FIT adder into a fixed FIT, but has done it so far only for rooftop solar PV (Kamolpanus, 2013). There is no local content requirement in Thailand’s renewable electricity support policies, but import duties create incentives to source locally (Beerepoot et al., 2013).46
46 These import duties were not considered in our analysis.
170 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries
3.4. Local and Global Learning in Renewable Energy in Thailand
As developed in Section 2, modern clean technology value chains feature significant local and global value creation. To illustrate how the local and global aspects of the value chain play out in Thailand, the value chain of a typical project in the electricity sector in Thailand is shown in Figure 4. Displayed is a value chain of a solar PV project, including the primary value chain, from material and component suppliers, up to the grid operator, and secondary activities such as universities, consultancies or legal services. For one specific project, a 9.5 MW solar PV project in Mae Chan in Chiang Rai province developed in 2013, we identified the most likely countries of origin for each value chain step. In the depicted case the project operator, the grid operator, the construction company, one of the two project developers, (probably) legal and financial services, and the regulator are local, while no core hardware components were manufactured in Thailand. The modules are manufactured by a Norwegian company in Singapore, while the inverters are made by a Swiss-Swedish company, most likely in Estonia. The leading production equipment suppliers, material suppliers and research institutes are located in Europe, the United States and Australia, thus it is almost certain that all these countries/continents are also represented in the value chain. The economics of the final project are determined by progress by all these actors in a concurrence of local and global learning effects, which calls for policy support that strengthens learning conditions locally and globally to facilitate overall technological progress.
171
Figure 4: VValue chain for an exemplary solar PV project in Thailand. Country codes as defined by UN Statistics Division; EU stands for European Union. Companies were identified from news sources.
4. Materials and Methods
4.1. General Model Framework
We developed a model of Thailand’s electricity sector and used different scennarios to estimate the effects of technological learning on the cost of achieving the Alternative Energy Development Plan (AEDP) targets. We chose a bottom-up, techno-economic model (Berglund and Söderholm, 2006) because it allows us to study the effects of cost dynamics on the technology-level on the aggregate cost of renewable energy policies (Kahouli-Brahmi, 2008). To model the effects of leearning on the cost of technologies, we chose the learning curve approach because it enables us to treat local and global effects separately (Hayward and Graham, 2013). We foocused on six renewable energy technologies under the AEDP: biomass, biogas, micro hydro, on-shore wind, solar PV, and concentrating solar power (CSP).
172 The Effect of Local and Global Learning on the Cost of Renewable Energy in Developing Countries
Model structure Key relationships between input and output variables (function of technology i, year t)
Globally installed Locally installed Learning rates (i) capacity (i, t) capacity (i, t)
Resource potential / Capital and O&M Debt & equity rates Tax and depreciation Fuel Cost (i, t) full-load hours (i, t) cost (i, t) (i) rules
Carbon credits Cost of generation by LCOE (i, t) Technology level (i,t)* source (i, t)
Sector level Total cost of generation (t)
Input / assumptions Incremental cost of Cost of avoided Fossil plant pipeline policy (t) generation (t) (t) Intermediate outputs Mitigation cost of Carbon emissions (t) Fuel mix (t) Output metrics policy (t)
Figure 5: Relationships between key input and output metrics in the techno-economic model; the upper half of the graphic shows variables calculated on the technology-level; the lower half shows variables calculated for the entire electricity sector/policy.
The overall structure of the model with its key variables and relationships is depicted in Figure 5. Calculating the cost of renewable electricity is a well-established process in renewable energy policy analysis (Burtraw et al., 2012). The cost of the avoided electricity, however, even though at least equally important, is often neglected (Schmidt et al., 2012). To obtain the cost of avoided electricity and the avoided greenhouse gas emissions, we compared different scenarios for diffusion of renewable energy with a hypothetical scenario without any renewables diffusion. Based on this comparison, the model provides two main outcome metrics, shown in dark grey in Figure 5, to assess the policy support needed to achieve Thailand’s renewable electricity targets: the incremental policy cost and the mitigation cost, both stated as net present value.47 The former is a proxy for marginal social cost (Palmer and Burtraw, 2005), while the latter allows comparisons between different mitigation measures and carbon prices.
How the scenarios deviate from the non-renewables scenario is calculated on the sectoral level. The incremental costs represent the difference between the total cost of renewable electricity and the total
47 The incremental costs and the mitigation costs are discounted to the year 2012 with the yield of 40-year Thai government bonds, which reflects the refinancing cost of the Thai government over the period of the assumed feed-in Tariff payments.
173
cost of avoided electricity (Schmidt et al., 2012). The mitigation costs are calculated based on the total carbon emission reductions from the AEDP and the incremental cost: